Agents and Multi-agent Systems: Technologies and Applications 2023: Proceedings of 17th KES International Conference, KES-AMSTA 2023, June 2023 9789819930678, 9789819930685

Citation preview

Smart Innovation, Systems and Technologies 354

Gordan Jezic · J. Chen-Burger · M. Kusek · R. Sperka · R. J. Howlett · Lakhmi C. Jain   Editors

Agents and Multi-agent Systems: Technologies and Applications 2023 Proceedings of 17th KES International Conference, KES-AMSTA 2023, June 2023

123

Smart Innovation, Systems and Technologies Volume 354

Series Editors Robert J. Howlett, KES International Research, Shoreham-by-Sea, UK Lakhmi C. Jain, KES International, Shoreham-by-Sea, UK

The Smart Innovation, Systems and Technologies book series encompasses the topics of knowledge, intelligence, innovation and sustainability. The aim of the series is to make available a platform for the publication of books on all aspects of single and multi-disciplinary research on these themes in order to make the latest results available in a readily-accessible form. Volumes on interdisciplinary research combining two or more of these areas is particularly sought. The series covers systems and paradigms that employ knowledge and intelligence in a broad sense. Its scope is systems having embedded knowledge and intelligence, which may be applied to the solution of world problems in industry, the environment and the community. It also focusses on the knowledge-transfer methodologies and innovation strategies employed to make this happen effectively. The combination of intelligent systems tools and a broad range of applications introduces a need for a synergy of disciplines from science, technology, business and the humanities. The series will include conference proceedings, edited collections, monographs, handbooks, reference books, and other relevant types of book in areas of science and technology where smart systems and technologies can offer innovative solutions. High quality content is an essential feature for all book proposals accepted for the series. It is expected that editors of all accepted volumes will ensure that contributions are subjected to an appropriate level of reviewing process and adhere to KES quality principles. Indexed by SCOPUS, EI Compendex, INSPEC, WTI Frankfurt eG, zbMATH, Japanese Science and Technology Agency (JST), SCImago, DBLP. All books published in the series are submitted for consideration in Web of Science.

Gordan Jezic · J. Chen-Burger · M. Kusek · R. Sperka · R. J. Howlett · Lakhmi C. Jain Editors

Agents and Multi-agent Systems: Technologies and Applications 2023 Proceedings of 17th KES International Conference, KES-AMSTA 2023, June 2023

Editors Gordan Jezic Faculty of Electrical Engineering and Computing University of Zagreb Zagreb, Croatia

J. Chen-Burger School of Mathematical and Computer Sciences The Heriot-Watt University Edinburgh, UK

M. Kusek Faculty of Electrical Engineering and Computing University of Zagreb Zagreb, Croatia

R. Sperka Department of Business Economics and Management Silesian University in Opava Karvina, Czech Republic

R. J. Howlett KES International Selby, North Yorkshire, UK

Lakhmi C. Jain KES International Selby, North Yorkshire, UK

Aurel Vlaicu University of Arad Arad, Romania Bournemouth University Poole, UK

ISSN 2190-3018 ISSN 2190-3026 (electronic) Smart Innovation, Systems and Technologies ISBN 978-981-99-3067-8 ISBN 978-981-99-3068-5 (eBook) https://doi.org/10.1007/978-981-99-3068-5 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

KES-AMSTA-2023 Conference Organization

KES-AMSTA-2023 was organized by KES International—Innovation in Knowledge-Based and Intelligent Engineering Systems.

Honorary Co-chairs I. Lovrek, University of Zagreb, Croatia Lakhmi C. Jain, KES International, Selby, UK

General Co-chairs Gordan Jezic, University of Zagreb, Croatia J. Chen-Burger, Heriot-Watt University, Scotland, UK

Executive Chair R. J. Howlett, Aurel Vlaicu University of Arad, Romania and Bournemouth University, UK

Programme Co-chairs M. Kusek, University of Zagreb, Croatia R. Sperka, Silesian University in Opava, Czech Republic

v

vi

KES-AMSTA-2023 Conference Organization

Publicity Co-chairs P. Skocir, University of Zagreb, Croatia M. Halaska, Silesian University in Opava, Czech Republic

International Programme Committee Assoc. Prof. Patricia Anthony, Lincoln University, New Zealand Dr. Alanis Arnulfo G., Instituto Tecnologico de Tijuana, Mexico Prof. Ahmad Taher Azar, Prince Sultan University, Saudi Arabia Assoc. Prof. Marina Bagi´c Babac, University of Zagreb, Croatia Prof. Dariusz Barbucha, Gdynia Maritime University, Poland Dr. Olfa Belkahla Driss, ESC University of Manouba, Tunisia Prof. Monica Bianchini, University di Siena, Italy Assoc. Prof. Bruno Blaskovic, University of Zagreb, Croatia Assoc. Prof. Frantisek Capkovic, Slovak Academy of Sciences, Slovakia Prof. Zeljka Car, University of Zagreb, Croatia Prof. Matteo Cristani, University of Verona, Italy Dr. Houssem Eddine Nouri, University of Gabes, Tunisia Prof. Margarita N. Favorskaya, Reshetnev Siberian State University of Science and Technology, Russia dipl.ing. Frano Skopljanac-Macina, University of Zagreb, Croatia Prof. Paulina Golinska-Dawson, Poznan University of Technology, Poland Dr. Michal Halaska, Silesian University in Opava, Czech Republic Dr. Madiha Harrabi, Esprit School of Engineering, Tunisia Prof. Tzung-Pei Hong, National University of Kaohsiung, Taiwan Prof. Dragan Jevtic, University of Zagreb, Croatia Prof. Arkadiusz Kawa, Pozna´n Institute of Technology, Poland Prof. Petros Kefalas, CITY College, University of York Europe Campus, Greece Prof. Zdenko Kovacic, University of Zagreb, Croatia Prof. Adrianna Kozierkiewicz, Wrocław University of Science and Technology, Poland Assoc. Prof. Konrad Kulakowski, AGH University of Science and Technology, Poland Prof. Setsuya Kurahashi, University of Tsukuba, Japan Prof. Mario Kusek, University of Zagreb, Croatia Prof. Kazuhiro Kuwabara, Ritsumeikan University, Japan Dr. Marin Lujak, University Rey Juan Carlos, Spain Prof. Rene Mandiau, University Polytechnique Hauts-de-France, France Dr. Bilel Marzouki, Esprit School of Business, Tunisia Prof. Manuel Mazzara, Innopolis University, Russia Prof. Jose M. Molina, Universidad Carlos III de Madrid, Spain

KES-AMSTA-2023 Conference Organization

vii

Prof. Radu-Emil Precup, Politehnica University of Timisoara, Romania Assist. Prof. Pavle Skoˇcir, University of Zagreb, Croatia Prof. Roman Sperka, Silesian University in Opava, Czech Republic Prof. Ryszard Tadeusiewicz, AGH University of Science and Technology, Poland Prof. Hiroshi Takahashi, Keio Unviersity, Japan Prof. Masakazu Takahashi, Yamaguchi University, Japan Prof. Takao Terano, Chiba University of Commerce, Japan Prof. Taketoshi Ushiama, Kyushu University, Japan Prof. Marin Vukovic, University of Zagreb, Croatia Assist. Prof. Ivona Zakarija, University of Dubrovnik, Croatia Dr. Mahdi Zargayouna, Université Gustave Eiffel, France

Invited Session Chairs Business Economics and Agent-Based Modelling Prof. Hiroshi Takahashi, Keio University, Japan Prof. Setsuya Kurahashi, University of Tsukuba, Japan Prof. Takao Terano, Chiba University of Commerce, Japan

Agent-Based Modelling and Simulation (ABMS) Assoc. Prof. Roman Šperka, Silesian University in Opava, Czech Republic

Education in Software Engineering in the Post-COVID Era Prof. Jean-Michel Bruel, Toulouse University, France Assoc. Prof. Salvatore Distefano, University of Messina, Italy Prof. Evgenii Bobrov, Innopolis University, Russia Mr. Nursultan Askarbekuly, Innopolis University, Russia Mr. Petr Zhdanov, Innopolis University, Russia Prof. Manuel Mazzara, Innopolis University, Russia Mr. Muhammad Ahmad, National University of Computer and Emerging Sciences, Pakistan

viii

KES-AMSTA-2023 Conference Organization

Intelligent Agents in Health, Wellness and Human Development Environments Applied to Health and Medicine Prof. Rosario Baltazar Flores, Tecnologico Nacional de Mexico—Campus Leon, Mexico Prof. Arnulfo Alanis Garza, Tecnologico Nacional de Mexico—Campus Tijuana, Mexico

Preface

This volume contains the proceedings of the 17th KES Conference on Agent and Multi-Agent Systems—Technologies and Applications (KES-AMSTA 2023) held in Rome, Italy, between 14 and 16 June 2023. The conference was organized by the KES International, its specialism group on agent and multi-agent systems, and the University of Zagreb, Faculty of Electrical Engineering and Computing. The KES-AMSTA conference is subseries of the KES Conference series. Following the success of KES Conferences on Agent and Multi-Agent Systems— Technologies and Applications, previously held in Rhodes, St. Julians, Gold Coast, Vilamoura, Puerto de la Cruz, Sorrento, Chania, Hue, Dubrovnik, Manchester, Gdynia, Uppsala, Incheon and Wroclaw, the conference featured the usual keynote talks, presentations and invited sessions closely aligned to its established themes. KES-AMSTA is an international scientific conference for discussing and publishing innovative research in the field of agent and multi-agent systems and technologies applicable in the digital and knowledge economy. The aim of the conference is to provide an internationally respected forum for both the research and industrial communities on their latest work on innovative technologies and applications that is potentially disruptive to industries. Current topics of research in the field include technologies in the area of decision making, big data analysis, cloud computing, Internet of things (IoT), business informatics, artificial intelligence, social systems, health, transportation systems and smart environments, etc. Special attention is paid on the feature topics of multi-agent systems and architectures, modelling and simulation of agents, intelligent business systems, agent negotiation and optimization, and intelligent agents applied to health and medicine. The conference attracted a substantial number of researchers and practitioners from all over the world who submitted their papers to the main track covering the methodologies of agent and multi-agent systems applicable in transportation, smart environments, digital and knowledge economy, and four invited sessions on specific topics within the field. Submissions came from 15 countries. Each paper was peerreviewed by at least two members of the International Programme Committee and International Reviewer Board. Thiry-nine papers were selected for presentation and publication in the volume of the KES-AMSTA 2023 proceedings. ix

x

Preface

The Programme Committee defined the main track multi-agent systems and the following invited sessions: agent-based modelling and transportation, business economics, intelligent agents in health, wellness and human development environments applied to health and medicine, business economics and education in software engineering in the post-COVID era. Accepted and presented papers highlight new trends and challenges in agent and multi-agent research. We hope that these results will be of value to the research community working in the fields of artificial intelligence, collective computational intelligence, health, robotics, smart systems and, in particular, agent and multi-agent systems, technologies, tools and applications. The chairs’ special thanks go to the following special session organizers: Prof. Rosario Baltazar Flores, Tecnologico Nacional de Mexico/Campus Leon, Mexico, Prof. Arnulfo Alanis Garza, Tecnologico Nacional de Mexico/Campus Tijuana, Mexico, Prof. Hiroshi Takahashi, Keio University, Japan, Prof. Setsuya Kurahashi, University of Tsukuba, Japan, Prof. Takao Terano, Chiba University of Commerce, Japan, Prof. Jean-Michel Bruel, Toulouse University, France, Assoc. Prof. Salvatore Distefano, University of Messina, Italy, Prof. Manuel Mazzara, Prof. Evgenii Bobrov, Mr. Petr Zhdanov and Mr. Nursultan Askarbekuly from Innopolis University, Russia, and Mr. Muhammad Ahmad, National University of Computer and Emerging Sciences, Pakistan, for their excellent work. Thanks are due to the programme co-chairs, all the programme and reviewer committee members, and all the additional reviewers for their valuable efforts in the review process, which helped us to guarantee the highest quality of selected papers for the conference. We cordially thank all the authors for their valuable contributions and all of the other participants in this conference. The conference would not be possible without their support. Zagreb, Croatia Edinburgh, UK Zagreb, Croatia Karvina, Czech Republic Selby, UK/Arad, Romania/Poole, UK Selby, UK April 2023

Gordan Jezic J. Chen-Burger M. Kusek R. Sperka R. J. Howlett Lakhmi C. Jain

Contents

Multi-agent Systems Agent-Based Medical Image Processing Using Multi-stage Distributed Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ˇ Armin Stranjak, Swen Campagna, and Igor Cavrak RoboTwin: Combining Digital Twin and Artificial Intelligence Domains for Controlling Robots in Industry 4.0 . . . . . . . . . . . . . . . . . . . . . . Flavien Balbo, Alaa Daoud, Guillaume Muller, Mihaela Juganaru-Mathieu, Fabien Badeig, Hiba Alqasir, and Mireille Batton-Hubert

3

15

A Method of Positioning a Humanoid Robot Relative to the Center of a Group of People—Analysis and Implementation . . . . . . . . . . . . . . . . . . Marko Kovaˇcevi´c and Zdenko Kovaˇci´c

27

How Can Adults Make Time to Study: A System for Employee Sharing and Reskilling Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kenta Abe, Minoru Matsui, and Hisashi Hayashi

41

Data Discretization for Data Stream Mining . . . . . . . . . . . . . . . . . . . . . . . . . Anis Cherfi and Kaouther Nouira Linear Machine Learning Algorithm for Early Annual Corn Yield Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ivan Kralj, Mario Kusek, and Gordan Jezic

55

65

Agent-Based Modelling and Transportation Multi-agent Modal Logic Evaluating Implicit Information . . . . . . . . . . . . Vladimir V. Rybakov

79

xi

xii

Contents

Multisectoral Household Location Agent-Based Simulation for Testing Policy Decisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simon Gorecki, Seghir Zerguini, Natalie Gaussier, and Mamadou Kaba Traore DaFne: Data Fusion Generator and Synthetic Data Generation for Cities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ayse Glass, Kübra Tokuç, Jörg Rainer Noennig, Ulrike Steffens, and Burak Bek

87

99

Application of TDABC Systems and Their Support with ABMS Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Michal Halaška and Roman Šperka Assessing the Impact of Shared-Taxi Pricing on Congestion Using Agent-Based Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 Negin Alisoltani and Mahdi Zargayouna Distributed, Classical and Flexible Job Shop Scheduling Problem with Transportation Times: A State-of-the-Art . . . . . . . . . . . . . . . . . . . . . . . 129 Bilel Marzouki, Olfa Belkahla Driss, and Khaled Ghedira Business Economics Proposal of Bicycle Sharing Operation System by Multi-agent Reinforcement Learning Using Transfer Learning . . . . . . . . . . . . . . . . . . . . 141 Kohei Yashima and Setsuya Kurahashi Change in Centrality and Team Performance: Inverse Relation Between Manager and Non-manager Communication . . . . . . . . . . . . . . . . 151 Hitomi Inagaki and Setsuya Kurahashi Digital Transformation and Cyber Threats for Small and Medium Sized Enterprises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Ralf-Christian Härting, Greg-Norman Schulz, Demian Deffner, and Christoph Karg Impact of Telework on Employee Satisfaction During the COVID-19 Crisis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Festina Kadriu, Julia Kleemann, Natalie Sorg, Ralf Härting, and Christopher Reichstein Bitcoin Fraudulent Transaction Detection Vulnerability . . . . . . . . . . . . . . . 183 Takashi Ehara and Hiroshi Takahashi The Relationship Between Technological Distance and Innovation Emergence in M&A Through Patent Data with Outlier Detection Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 Daishiro Yamamoto and Hiroshi Takahashi

Contents

xiii

Estimated Economic Ripple Effects of Closed-Type Tourism Promotion Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 Yoko Ishino and Hideto Nakamura Intelligent Agents in Health, Wellness and Human Development Environments Applied to Health and Medicine Analysis of Batch Size in the Assessment of Bone Metastasis from Bone Scans in Various Convolutional Neural Networks . . . . . . . . . . 221 Vincent Peter C. Magboo and Patricia Angela R. Abu Exploratory Study of Eye-Tracking Path: A Case Study from Switzerland . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Kholod Marina, Wirth Julia, El Darawany Ahmed, and Shilina Marina Patient Monitoring Through Intelligent Agents: A Preliminary Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 Ángeles Arellano Vera, Rosario Baltazar, J. Ascención Guerrero-Viramontes, Arnulfo Alanis, Juan José Soto-Bernal, and R. González-Mota Designing a BCI Platform with Embedded ANN as an Aid for Autism Spectrum Disorder (ASD) Diagnosis: A Preliminary Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 Gerardo Vilchis, Rosario Baltazar, Arnulfo Alanis, J. Francisco-Mosiño, and Miguel Angel Casillas-Araiza Education in Software Engineering in the Post-COVID Era Cultural-Ethical Evaluation in the Launch of AI Education Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 Sergei Kladko An Analytic Hierarchy Process for Student Motivation . . . . . . . . . . . . . . . . 277 Robiul Islam, Leonard Johard, Mohammad Reza Bahrami, and Mustafeed Zaman Training Students as Agile Developers: Team and Role Building Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289 Paolo Ciancarini and Marcello Missiroli Impact of Covid-19 on Employee Satisfaction and Trust with Focus on Working from Home . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301 Miriam Gazem, Ralf Härting, Anna Schneider, and Christopher Reichstein An Approach for Organising and Managing an Academic Year Using Online Tools and Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313 Liviu-Andrei Scutelnicu and Marius Ciprian Ceobanu

xiv

Contents

Exploring the Impact of COVID-19 on Education: A Study on Challenges and Opportunities in Online Learning . . . . . . . . . . . . . . . . . 323 Ananga Thapaliya and Yury Hrytsuk Memes as a Memorization Technique in Education . . . . . . . . . . . . . . . . . . . 333 Hamza Salem and Siham Siham Hattab Quantifying Education in the Post-COVID Era: An Engineering Approach Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339 Gerald B. Imbugwa and Tom Gilb Teaching Object-Oriented Requirements Techniques: An Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 Maria Naumcheva Onlife Education: Beyond Distance Learning by Intelligent Tutoring Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355 Salvatore Distefano A Personal View on Past and Future Higher Education . . . . . . . . . . . . . . . 367 Nikola Zlatanov Running Regular Research Seminar Online . . . . . . . . . . . . . . . . . . . . . . . . . . 373 N. V. Shilov, D. A. Kondratyev, N. Kudasov, and I. S. Anureev Innopolis University: An Agile and Resilient Academic Institution Navigating the Rocky Waters of the COVID-19 Pandemic . . . . . . . . . . . . . 383 Yuliya Krasylnykova, Iouri Kotorov, Jaroslav Demel, Manuel Mazzara, and Evgeny Bobrov Teaching the Future: The Vision of AI/ChatGPT in Education . . . . . . . . . 393 Mohammad Reza Bahrami, Bahareh Bahrami, Farima Behboodi, and Samae Pourrafie Thesis Supervision in Computer Science—Challenges and a Gamified Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403 Rabab Marouf, Iouri Kotorov, Hamna Aslam, Yuliya Krasylnykova, and Marko Pezer Humanizing Zoom: Lessons from Higher Education in Qatar . . . . . . . . . . 413 R. Bianchi, B. Yyelland, A. Weber, K. Kittaneh, Sara Mohammed, Aia Zaina, Afreena Niaz, Huda Muazzam, Selma Fejzullaj, and Lolwa Al-Thani Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423

About the Editors

Gordan Jezic is a professor at the University of Zagreb, Croatia. His research interest includes telecommunication networks and services focusing particularly on parallel and distributed systems, machine-to-machine (M2M) and Internet of things (IoT) systems, communication networks and protocols, mobile software agents and multiagent systems. He actively participates in numerous international conferences as a paper author, speaker, member of organizing and programme committees or reviewer. He co-authored over 100 scientific and professional papers, chapters and articles in journals and conference proceedings. J. Chen-Burger is an assistant professor at the department of Computer Science at Heriot-Watt University. She was a research fellow of Informatics at the University of Edinburgh. Her research interests include enterprise modelling, process modelling, execution and mining technologies and how they may interact with agent technologies to solve complex real-world problems. She is a committee member of several international conferences and journals and the chair of conference and conference sessions. She is PI to several research and commercial projects. M. Kusek is a professor at the University of Zagreb, Croatia. He holds Ph.D. (2005) in electrical engineering, from the University of Zagreb. He is currently a lecturer of 9 courses and has supervised over 130 students at B.Sc., M.Sc. and Ph.D. studies. He participated in numerous projects in local and internationals. He has co-authored over 80 papers in journals, conferences and books in the area of distributed systems, multiagent systems, self-organized systems and machine-to-machine (M2M) communications. He is a member of IEEE, KES International and the European Telecommunications Standards Institute (ETSI). He serves as a programme co-chair at two international conferences. R. Sperka is an associate professor and the head of Department of Business Economics and Management at Silesian University in Opava, School of Business Administration in Karvina, Czech Republic. He holds Ph.D. title in “Business economics and management” and Dr. title in “Applied informatics” since 2013. He xv

xvi

About the Editors

has been participating as a head researcher or a research team member in several projects funded by Silesian University Grant System or EU funds. His field of expertise is business process management, process mining, implementation and deployment of information systems and software frameworks; the use of agent-based technology in social sciences; and modelling and simulation in economic systems and financial markets. Dr. R. J. Howlett is the executive chair of KES International, a non-profit organization that facilitates knowledge transfer and the dissemination of research results in areas including intelligent systems, sustainability and knowledge transfer. He is a visiting professor at Bournemouth University in the UK. His technical expertise is in the use of intelligent systems to solve industrial problems. He has been successful in applying artificial intelligence, machine learning and related technologies to sustainability and renewable energy systems; condition monitoring, diagnostic tools and systems; and automotive electronics and engine management systems. His current research work is focused on the use of smart microgrids to achieve reduced energy costs and lower carbon emissions in areas such as housing and protected horticulture. Prof. Lakhmi C. Jain, Ph.D., Dr. H.C., ME, BE (Hons), Fellow (Engineers Australia), is with the University of Arad. He was formerly with the University of Technology Sydney, the University of Canberra and Bournemouth University. Professor Jain founded the KES International for providing a professional community and the opportunities for publications, knowledge exchange, cooperation and teaming. Involving around 5000 researchers drawn from universities and companies worldwide, KES facilitates international cooperation and generates synergy in teaching and research. KES regularly provides networking opportunities for professional community through one of the largest conferences of its kind in the area of KES. His interests focus on the artificial intelligence paradigms and their applications in complex systems, security, e-education, e-health care, unmanned air vehicles and intelligent agents.

Multi-agent Systems

Agent-Based Medical Image Processing Using Multi-stage Distributed Neural Network ˇ Armin Stranjak, Swen Campagna, and Igor Cavrak

Abstract In this paper, we describe possible applications of early exit deep neural networks in magnetic resonance imaging, aiming to improve patient scan times and reduce processing costs. The solutions rely on deep neural network layers deployed over a hierarchical processing environment of embedded, edge-based and shared cloud-based resources. Coordination among processing environment layers is achieved by representing different processing resources as agents, effectively forming a multi-agent system. The described approach combines the least-required processing property of the early exit deep neural network and the resource usage optimisation of the multi-agent system in the dynamic and open environment.

1 Introduction Recent years have seen many success stories involving the use of advanced artificial intelligence (AI) in the healthcare industry, ranging from relatively straightforward patient monitoring systems and IoHT devices [7] to sophisticated algorithms for the medical imaging analysis and other cutting-edge medical examination techniques [6]. Deep neural networks (DNNs) [8], in particular, offer a promising AI method in complex medical analysis, but require a lot of processing power. As a solution, neural network partitioning techniques have been proposed [4] to distribute the computing effort among available computing resources. This paper focuses on the inference phase of early exit deep neural networks (eeDNN) used in Magnetic Resonance Imaging (MRI) systems. Such a technique is employed primarily to analyse resulting MRI slice images, but also in other pre- or post-examination MRI workflow stages. The goals of MRI workflow processing are A. Stranjak (B) · S. Campagna Siemens Healthcare GmbH, Allee am Röthelheimpark 2, 91052 Erlangen, Germany e-mail: [email protected] ˇ I. Cavrak Faculty of Electrical Engineering and Computing, University of Zagreb, Unska 3, 10000 Zagreb, Croatia © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_1

3

4

A. Stranjak et al.

two-fold; the central aspect is medical diagnostics, while the second is improving the utilisation of high-value medical equipment by maximising patient throughput. Using complex AI algorithms, such as DNN, on massive amounts of data generated by MRI workflow is difficult if the processing is limited to locally available computing power. Offloading processing to a remote, more powerful edge and cloudbased computing infrastructure necessitates high throughput connectivity and introduces network-related delays. Adding more computing power to the baseline MRI equipment would reduce local processing times but raise the cost of MRI equipment and introduce additional computing resources that would be unused most of the time. However, completely offloading the processing to a cloud-based infrastructure would result in longer wait times, lower patient throughput, and underutilisation of expensive MRI equipment. One of the solutions lies in the distribution of required processing over different layers of the scalable distributed computing hierarchy [5], where different segments of a DNN are distributed over heterogeneous processing nodes. In addition, the early exit technique can be used [14], where the early exit distributed neural network (eeDNN) processing is propagated to higher eeDNN layers only if non-satisfactory results are achieved in lower eeDNN layers. A multi-agent system is employed to select the optimal processing resources in distributed computing hierarchy. Through negotiation among agents, the system effectively maps one or more stages of an eeDNN to available computing resources in the cloud layer. Thus, the system tries to achieve the lowest possible processing times that satisfy the defined result quality, facilitating, in most cases, faster patient processing and increasing MRI equipment utilisation while requiring only modest local computing resources on the MRI equipment. The remainder of the paper is organised as follows. Section 2 describes the specifics of the MR Imaging workflow, and Sect. 3 presents a brief overview of the usage of multi-agent systems in the MR Imaging domain. Section 4 introduces the proposed MAS model combining eeDNN and multi-agent system into a coherent resource-negotiation framework. Section 5 defines the MAS negotiation protocol and award function. Finally, Sect. 6 concludes the paper.

2 Application in Magnetic Resonance Imaging Workflow The workflow of Magnetic Resonance (MR) Imaging systems is typically relatively complex compared to other medical imaging technologies due to the high flexibility and some intrinsic properties of Magnetic Resonance Imaging principles. Additionally, Magnetic Resonance Imaging systems have significant computing power requirements when using state-of-the-art algorithms, both for executing single measurements to compose a complete MR examination and to support and ease the workflow of an entire examination for the operator of the system. In this section, we will highlight some use cases where we envision that our proposed strategy of eeDNN can be applied directly to the major processing problem of Magnetic Resonance Imaging:

Agent-Based Medical Image Processing Using Multi-stage …

5

insufficient local processing power for image acquisition and reconstruction and the consequent high cost of the offered MR system. In our proposal, only computationally “cheap” neural network computation needs to be calculated at the MR system itself, and only in case the locally calculated result is not of sufficient quality, further refinement of the calculation is offloaded to more powerful, cloud-based processing units. Every MR examination starts with positioning a patient on the patient table. This includes several work steps like precisely placing the receive coils to acquire the magnetic resonance signals, providing hearing protection means and several more. In order to use the full potential of the MR system during measurements, the exact patient position and orientation, as well as patient parameters like weight and size, are crucial. 3D cameras on the ceiling can acquire precise data about the patient before being transferred into the bore of the MR system. With the help of novel AI algorithms, the required parameters can be derived from this data and by using our proposed eeDNN even on MR systems with limited local computing power. The patient examination starts with a measurement generating overview images. Based on this overview, the exact position of the subsequent measurements can be determined, followed by the image acquisition of the intervertebral discs of the lumbal spine, for example. In fact, one must acquire an overview image of the lumbal spine, and based on this measurement, precise instruction is necessary on where exactly to acquire the image slices of the individual intervertebral discs. State-of-the-art AI algorithms already automate this task, and with the assistance of eeDNN, such processing can also be achieved with limited local computing power. Another important step during an MR examination is the image quality (IQ) assessment of the acquired images, if the image acquisition was successful. For example, the patient’s body moving during a measurement, even for a short period of time, will result in poor image quality and will not be suitable for a subsequent diagnosis. Thus, it is essential to immediately perform a quality assessment as long as the patient is still in the MR system so that either the measurement can be repeated or other measurement strategies to be considered. Equally, the IQ assessment can be automated by state-of-the-art AI algorithms, and their use can be enabled even using low computing power systems with the help of the proposed eeDNN. Finally, automated organ segmentation is a diverse area of algorithms to do additional post-processing of the acquired images before doing a subsequent reading of the results. AI algorithms can be used to achieve this goal, and our proposed eeDNN helps to execute them even on MR systems with limited local computing power.

3 Multi-agent Systems in Medical Imaging Computational and technological advancements through the utilisation of multiagent systems have been proven by various success stories, including automatic negotiations and game theory, planning and scheduling services, application in autonomous vehicles, robotic swarms, etc. In recent years, multi-agent systems have

6

A. Stranjak et al.

started to make significant gains in the disciplines like medical sciences and AI-based diagnostics. Medical Image Segmentation and image reading [3] utilised a multiagent approach to image segmentation to extract objects of interest and coordinate the outcome of the agents’ work to construct an anatomical structure. Similarly, a significant set of e-Health initiatives and projects [2] are emersed into content-rich, decentralised, interdependent dynamic environments, which are found appropriate for using multi-agent systems. An image interpretation difficulty, especially a sheer number of image readings required by the radiology department, is addressed through the usage of multi-agent systems [1]. We have also addressed in Chap. 2 several use cases within the MR image analysis and processing. These activities rarely demand only simplified, centralised processing where the image analysis is non-time critical and quality tolerant. However, a significant improvement in image quality and more complex algorithms for image-related task processing required an increased computing power and innovative approach. We have addressed this challenge in our previous work [10]. Such a demanding environment led further to the consideration of neural network processing. If fused with the higher demand for computational processing, it provides a contextual suitability for multi-agent systems and cloud-based computing. This paper addresses such fusion.

4 Model In our processing model, we adopt the baseline eeDNN model [14], with deep neural network layers partitioned into three stages—local, edge and cloud (Fig. 1, left). Early

Fig. 1 eeDNN structure (left) and network stage placement scenarios for three-stage networks and local-edge-cloud hierarchical computing environment (right)

Agent-Based Medical Image Processing Using Multi-stage …

7

exit layers (local and edge) perform final processing at the corresponding stage and permit quality assessment of the achieved result (classification, segmentation etc.), allowing the system to decide whether to conclude the processing or advance it to the upper DNN stage. As a rule of thumb, the number of DNN layers in a stage corresponds to the expected processing power in the corresponding layer of the hierarchical computing environment, with the local stage expected to be the least powerful one. Result quality assessment is based on the entropy of the achieved result [13], where entropy lower than the threshold value implies early exit result being treated as the final processing result and returned to the user. In the opposite case, the processing results of the uppermost DNN layer of the current stage are transferred to the input of the first DNN layer of the higher stage, where the processing is resumed. Benefits of the early exit mechanism are three-fold: (i) eeDNNs inherently possess the least-required-processing property: data transfer to and higher-stage processing is performed only if required result quality cannot be achieved at the lower stage; (ii) significant data compression is achieved when transferring processing results between DNN stages, as opposed to the raw MRI input; and (iii) system resilience—unavailability of higher DNN stages does not automatically imply service unavailability—but may yield lower quality results. Different static or dynamic placement strategies are possible concerning the threestaged structure of the eeDNN (Fig. 1, right). Scenario #1 placement implies mapping each eeDNN stage to a separate distributed hierarchical computing layer, where the local stage is being allocated to the locally available computing power, the more demanding stage is deployed at the edge layer, and the highest stage is deployed at the cloud layer. The second scenario implies co-location of edge and cloud stages in case of usage scenario where edge layer resources are not available. Finally, the third scenario implies the availability of cloud-only processing resources, where raw input data is transferred to the cloud, averaging the highest communication delays and dependence on stable data links. The model proposed in this paper focuses only on Scenario #1, where there is a 1:1 mapping between an edge resource and the local resource, and there exists a dynamic marketplace of cloud-based resources. The placement of three eeDNN stages in this scenario implies fixed mapping of local and edge stages to known computing resources, but the cloud stage-to-resource mapping is done dynamically, using a multi-agent system and a bidding process. Fusion between eeDNNs and the open environment of multi-agent systems proved to be a challenging task. Tight coupling of disparate layers of DNN networks requires strong binding between bordering neurons to encompass a holistic and consistent execution environment for neural processing. On the other hand, market-based systems enable a flexible and open environment where offered services are validated through negotiation and competition mechanisms while service agents arbitrarily choose when and how to provide their processing activities. Therefore, we propose to distribute three distinct layers of processing (Fig. 2), determined and shaped by the heterogeneous character of every layer found in the hierarchical distributed computing model [5].

8

A. Stranjak et al.

Fig. 2 The contextual differences of hierarchical neural network layers

The first level of processing, normally associated with the embedded or limited local power, is organised as a fixed, predefined network layer operating on locally sourced data. Neural network inputs are fed directly from the local imaging process and desired entropy level is targeted by best-effort approach. Typical local systems would be low-power computing nodes of a mobile MR scanner or an embedded camera with a localised embedded processing device. Had desired entropy of the first network level not been achieved, a dedicated proxy device is defined in the next, higher hierarchy layer. The second layer determines the role of a dedicated proxy device associated with the network nodes from the first layer. Since such device is dedicated to its own pairing device from the local environment, its network configuration is fixed and determined in advance. In case of an unsatisfactory entropy level achieved by the first layer, a vector of weight values is transported as an input to the second layer. Since the achieved entropy level in the first layer is dependent on several factors, including input data complexity, desired quality output, appropriateness of chosen network configuration and so on, the data transferred with weight vector also includes a desired entropy level. This value will be used in comparison with the entropy level of the second layer output of the neural network. In case that desired entropy level is achieved, the results will be provided back to the first layer without further involvement of the third layer.

Agent-Based Medical Image Processing Using Multi-stage …

9

The third layer brings wider flexibility in terms of connectivity and processing possibilities. Instead of adopting another peer-to-peer connection characteristics, like a link between the first and second layer, a more fitting solution would be a market-based negotiation environment. Processing nodes are represented by cloudbased agents as autonomous entities. They would engage in an open negotiation with the Proxy devices, who are also equipped with a multi-agent communication stack, while competing with other cloud agents for processing services. Proxy device would collaborate with the cloud agents in order to agree and achieve fast execution of the final network layer. Unlike in the case between the first and second layer, where expected entropy is imposed on Proxy device, cloud agents are not tasked with such demand as they will inherently aim for the best possible result, as agreed during the negotiation conversation. However, our approach enables independent, commercially cloud-based entities to participate in such negotiation by offering their computer resources with the possibility to dynamically adopt a given network topology during negotiation by accessing such a topology configuration remotely.

5 Market-Based Environment for Multi-agent Systems Goal autonomy and independent decision-making are the fundamental characteristics of any multi-agent system. In order to ensure optimal or, in given circumstances, the best possible outcome, mutual competition is encouraged among participating agents. Since their own goals stay private, the establishment of design mechanisms to encourage agent to reveal their true intentions or estimations is important for robust and stable dynamic systems. It is, therefore, crucial to ensure an appropriate mechanism design to reveal the agent’s true preferences. Since the agents do not know their preferences in advance, they would require to gather the information before deliberation about the allocation of their resources and the next actions. This approach is especially beneficial, as will be shown later when the design of the negotiation protocol needs to encompass both: the task description for the agent to execute and the corresponding award as compensation for the successful computational performance. Various incentivisation strategies were developed to encourage the deliberative agents to reveal their true intentions. A conversation protocol needs to be designed in such a way that the deliberate untruthfulness of a particular participant would be fundamentally against his own interest or goals, and such behaviour would be seen as irrational. A reasonably simple solution would be based on Vickrey auction [15] where truthful revealment of the agent’s private intentions is achieved by committing the winning agent with the second-best offer. In such a way, the agents are incentivised not to speculate on the offers of other agents, but their truthfulness in their negotiation bids is their best possible strategy. For this reason, we would base our negotiation design on such an approach.

10

A. Stranjak et al.

Fig. 3 Negotiation protocol (left) and corresponding sequence diagram (right)

5.1 Scenario-Based Negotiation Negotiation protocols of multi-agent systems provide contextual and conversational boundaries within which the bidding process takes place. The main focus on implementation-independent scenario description, as well as a provision of a communication stack generation for transparent agent coordination, led us to propose the usage of SDLMAS platform. The platform offers a declarative and expressive domain-specific language for a scenario description, code generation of communication layers and provision of stub placeholders for the implementation of the agent’s business logic. Provided placeholders and generated conversation state machine enable the designer to concentrate on business logic implementation while message performative reception and transmission conditions are automatically generated. Further implementation details can be found in [11, 12]. Negotiation protocol description and corresponding sequence diagram are shown in Fig. 3 and is based on Contract-Net Protocol [9]. The description starts with the definition of all actors in the negotiation, namely Proxy and Processor agents. As pointed out earlier, the Proxy agent includes its own network layer in the calculation, and if the exit result entropy is not satisfying, as defined by the agreement with the first layer, the Proxy agent will initiate the

Agent-Based Medical Image Processing Using Multi-stage …

11

negotiation by sending a message with CFP performative to all Processor agents using declaration. The content of this “call for proposal” message contains the vector of weight values transported from the second layer, guaranteed value A0 if a task is executed at the time estimated by the Processor agent, and maximum award value Am in case that the processing result is provided earlier. The details about an award function are provided in Sect. 5.2. Based on these values, Processor agents reply with their proposals (message PROPOSE) providing estimated time Te when the processing will be completed. Alternatively, they can simply terminate the conversation by sending NO_PROPOSAL message. Within a given timeout, the Proxy agent collects all received offers and, based on the principles of the Vickrey auction, selects the best offer, but it offers the second-best estimated time back to the winning Processor agent. This OFFER message also contains the additional parameters Td , k and q that completes the definition of the award function. Td parameter defines the deadline for the reception of the processing results, after which no award will be provided. The remaining k and q parameters are used to define the “curvature” of the award function, giving a chance to the Proxy agent to shape the award function based on the urgency of processing. The winning Processor agent receives a complete set of parameters to define the award function. If it is satisfactory, it has a chance to accept or reject the offer (corresponding ACCEPT and REJECT messages are used) based on its estimation of award function and its cost-effectiveness in exchange for the processing time and result. If the offer is satisfactory, the Proxy agent receives the acceptance of the offer, and it still has a final chance to commit or decline the acceptance using COMMIT or DECLINE performatives. After the reception of the task commitment, the Processor agent performs the agreed task. The time difference between the transmission of the COMMIT message and the reception of the INFORM message by Proxy agent is used as reception time Tr which is used to calculate an award At that will be transferred to Processor agent by AWARD message. Otherwise, if the result is provided after the deadline time Td , no award is given, and NO_AWARD message is sent. This protocol does not dictate the exact award character, financial or symbolic, and one possible implementation could include some sort of cryptocurrency or locally sourced award tokens.

5.2 Award Function The award function and corresponding function definition are depicted at Fig. 4. As explained above, the winning agent is awarded based on an agreed award function shown above. Its parameters A0 , Am , Te and Td define the functional dependency between the time of delivered results and award amount. The choice of the exponential function (top function definition in Fig. 4) seems reasonable as most natural and market-based processes and bidding mechanisms could be accurately explained with natural exponential functions. In addition, k and q exponents that define the “curvature” of the award function are also agreed upon as a part of the negotia-

12

A. Stranjak et al.

Fig. 4 Award function diagram and the functional definition

tion conversation, as explained in 5.1. If we want to transform given coefficients into negotiation parameters mentioned above, we get the second form for the award function, shown as a bottom function definition on right in Fig. 4. Since two different exponential functions describe the award given before and after estimated time Te , the condition was that they provide the same value at the point (Te , A0 ) with the additional condition that the first function contains point (0, Am ) and the second one contains (Td , 0). Based on these conditions, the bottom award function was deducted from the top one.

6 Conclusion and Future Work In this paper, we proposed a multi-agent based framework for advanced data processing in Magnetic Resonance Image systems using early exit distributed neural networks. To coordinate the usage of available computing resources hosting layers of eeDNN within distributed computing hierarchy, we employ a multi-agent system, with self-interested agents representing computing resources in different layers. Through negotiation among agents, based on a reward mechanism, currently optimal computing resources are selected in different layers for processing data within eeDNN, effectively composing the deep neural network processing infrastructure on-demand. The future work will include the detailed study of optimal k and q parameters of the award function for various bidding strategies. Further interest will include a careful analysis of received results and the establishment of the trust mechanisms in the open, market environment where, in addition to agent’s processing capability, a trust parameter will be included in the negotiation protocols.

Agent-Based Medical Image Processing Using Multi-stage …

13

References 1. Alves, V., et al.: Medical imaging environment—a multi-agent system for a computer clustering based multi-display. In: Progress in Artificial Intelligence, pp. 332–343. Springer, Berlin (2007) 2. Bergenti, F., Poggi, A.: Multi-agent systems for e-health: recent projects and initiatives. 10th Workshop on Objects and Agents, WOA’09, 01 (2009) 3. Chitsaz, M., Seng, W.: Medical image segmentation using a multi-agent system. Int. Arab J. Inf. Technol. 10, 05 (2013) 4. Dean, J., et al.: Large scale distributed deep networks. In: Proceedings of the 25th International Conference on Neural Information Processing Systems, vol. 1, NIPS’12, pp. 1223–1231 (2012) 5. Skala, K., et al.: Scalable distributed computing hierarchy: cloud, fog and dew computing. Open J. Cloud Comput. (OJCC) 2(1), 16–24 (2015) 6. Lin, W., et al.: Convolutional neural networks-based MRI image analysis for the Alzheimer’s disease prediction from mild cognitive impairment. Front. Neurosci. 12, 777 (2018) 7. Rodrigues, J.J.P.C., et al.: Enabling technologies for the internet of health things. IEEE Access 6, 13129–13141 (2018) 8. Schmidhuber, J.: Deep learning in neural networks: an overview. Neural Networks 61, 85–117 (2015) 9. Smith: The contract net protocol: high-level communication and control in a distributed problem solver. IEEE Trans. Comput. C-29(12), 1104–1113 (1980) 10. Stranjak, A., Campagna, S.: Decentralised agent-based medical image reconstruction. Procedia Comput. Sci. 207, 2106–2115 (2022). Knowledge-Based and Intelligent Information & Engineering Systems: Proceedings of the 26th International Conference KES2022 ˇ 11. Stranjak, A., Cavrak, I., Žagar, M.: Scenario description language for multi-agent systems. In: New Frontiers in Applied Artificial Intelligence, pp. 855–864. Springer, Berlin (2008) ˇ 12. Stranjak, A., Cavrak, I., Žagar, M.: Scenario description language in multi-agent simulation system. In: Agent and Multi-agent Systems: Technologies and Applications, pp. 169–179. Springer, Berlin (2011) 13. Teerapittayanon, S., McDanel, B., Kung, H.T.: Branchynet: fast inference via early exiting from deep neural networks. In: 2016 23rd International Conference on Pattern Recognition (ICPR), pp. 2464–2469 (2016) 14. Teerapittayanon, S., McDanel, B., Kung, H.T.: Distributed deep neural networks over the cloud, the edge and end devices. In: 2017 IEEE 37th International Conference on Distributed Computing Systems (ICDCS), pp. 328–339 (2017) 15. Varian, H.R.: Economic mechanism design for computerized agents. In: Proceedings of the First USENIX Workshop on Electronic Commerce (1995)

RoboTwin: Combining Digital Twin and Artificial Intelligence Domains for Controlling Robots in Industry 4.0 Flavien Balbo, Alaa Daoud, Guillaume Muller, Mihaela Juganaru-Mathieu, Fabien Badeig, Hiba Alqasir, and Mireille Batton-Hubert

Abstract One of the main challenges in Industry 4.0 is the supervision and coordination of heterogeneous robots at runtime, especially when they have a certain level of autonomy, as seen in Autonomous Mobile Robots. In addition, autonomous robots and their digital twins are designed by private manufacturers, and their code is often inaccessible. In this study, we present a solution to anticipate and control the behavior of robots by combining Digital Twin (DT) and Artificial Intelligence (Multiagent System and Machine Learning (ML)) models and technologies in a nonintrusive way. Using DT technologies we reproduce the action model of the robot as an autonomous agent behavior, treating the robot and its DT as black boxes. We apply existing ML techniques on the logs of its actions in various situations to predict the robot actions in the near future. We illustrate the use of our framework with a conflict resolution situation for a new AMR in a factory.

F. Balbo (B) · A. Daoud · M. Batton-Hubert Mines Saint-Etienne, University of Clermont Auvergne, INP Clermont Auvergne, CNRS, UMR 6158 LIMOS, Saint-Etienne, France e-mail: [email protected] A. Daoud e-mail: [email protected] M. Batton-Hubert e-mail: [email protected] G. Muller · M. Juganaru-Mathieu · F. Badeig Mines Saint-Etienne, Institut Henri Fayol, Saint-Etienne, France e-mail: [email protected] M. Juganaru-Mathieu e-mail: [email protected] F. Badeig e-mail: [email protected] H. Alqasir Laboratoire Hubert Curien UMR 5516, 18 rue Benoit Lauras, 42000 Saint-Etienne, France © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_2

15

16

F. Balbo et al.

1 Introduction A challenge facing Industry 4.0 systems is to manage heterogeneous robots in a factory. It requires tools that are independent of the manufacturers for supervising and coordinating robots at runtime, especially when these entities have a high level of autonomy, as Autonomous Mobile Robots (AMR) [5]. Individual robots are designed to achieve their tasks, usually without exhibiting an explicit description of their behavior in different situations, i.e. what autonomous decision might it make corresponding to some event. To use them in different contexts and, even better, to improve the global system, it becomes necessary to model the way a given robot will act in a given environment. Designed by private manufacturers, these robots are often black-boxes and non intrusive solutions have to be found for achieving control. In this work, we propose a framework, called RoboTwin, consisting in a methodology and its tools combining Digital Twin (DT) and Artificial Intelligence (MultiAgent System (MAS) and Machine Learning (ML)) models and techniques for controlling a robot into an existing real system. We reproduce the action model of a robot as an autonomous agent behavior in order to control the robot. We aim to achieve this objective by applying existing machine learning techniques to predict the robot actions in the near future, given the logs of the actions performed in various situations collected by simulation of the robot digital twin. This paper is structured as follows: Sect. 2 overviews the background and related works; Sect. 3 presents our hypothesis; Sect. 4 details our methodology and tools; Sect. 5 illustrates our proposal with a proof of concept; Last section concludes and proposes research perspectives.

2 Background Our approach involves observing the behavior of a robot’s digital twin, learning its action model, abstracting this digital twin as an autonomous agent that adopts the learned action model, and using it to control the real robot. In this section, we provide an overview of the models and technologies used in RoboTwin.

2.1 Digital Twin (DT) According to the Gemini Principles [2], a digital twin is “A realistic digital representation of something physical.” What sets a digital twin apart from any other digital model is its connection to the physical twin [2]. The concept and terminology of the digital twin are growing in academia, and this growth is being enhanced by advances in the Internet of Things and Artificial Intelligence. While a set of application areas

RoboTwin: Combining Digital Twin and Artificial Intelligence …

17

for digital twins has been listed by [6], industry and manufacturing have taken the lion’s share of digital twin applications. There are common misconceptions in the literature about distinguishing digital twins from general computing models and simulations. Based on the existence and direction of data flows between the physical and digital entities, three concepts [6] are distinguished: digital model (no automatic data flow), digital shadow (physical-to-digital one-way data flow), and digital twin (bidirectional automatic data flow). The same authors argue that one of the common goals of digital twins is the predictability and learnability of the system. In this work, we fully adopt the Gemini Principles and the Fuller et al. point of view on DT: “A change made to the physical object automatically leads to a change in the digital object and vice versa.”

2.2 Machine Learning Approaches for Behavior Learning The problem of reproducing a behavior model, also known as action-model learning or behavior learning, has been addressed using various machine learning (ML) techniques. However, the literature lacks a comparison of these approaches. In this section, we provide a brief overview of different ML approaches used for behavior learning and highlight their strengths and limitations. Arora et al. [1] broadly classifies ML approaches to reproduce planning action models based on several criteria. They can be classified into online and offline methods. Online methods learn in real-time while interacting with the environment, while offline methods learn from a pre-collected dataset of action traces. In offline methods, the Learning by Observation in Planning Environments (LOPE) system [7] learns planning operators by observing the consequences of the execution of planned actions in the environment. On the other hand, the probabilistic planning operators are learned incrementally in [14]. A system that learns the effects of an agent’s action given a set of actions and their preconditions, in a partially observable and noisy environment was proposed in [12]. Recurrent Neural Networks (RNNs) are a well-known method for learning sequence models [11]. In the domain of activity prediction, several notable works use the version of RNNs called Long Short-Term Memory (LSTM). For example, [10] develop a human movement trajectory prediction system that incorporates scene information (Scene-LSTM). Similarly, [9] used LSTM to explore the movement patterns of different heterogeneous traffic agents and predict their future trajectories accurately to help the autonomous vehicle make reasonable navigation decisions. Tran et al. [15] studied the problem of forecasting human trajectory into the far future and show that explicitly characterizing the dependency of pedestrians’ future trajectory on their goal provides strong consistency in long-term prediction and reduces the progressive accumulation of error. In this work, we propose a general framework that leverages the strengths of the different ML approaches described above. Depending on the types of inputs/outputs and the pattern of acquisition of the data that will be used to train the ML model,

18

F. Balbo et al.

various approaches can be used. For instance, if we have an offline dataset with a sequence of previous actions as input and a discrete last action output, the ML model used can consist of RNNs. If we acquire data online, we can use Reinforcement Learning techniques.

2.3 MultiAgent System (MAS) Prototyping and abstracting Multiagent Systems (MAS) paradigm can be a suitable approach to addressing the challenges of prototyping and abstracting Multi-Robot Systems (MRS). According to [8], the main differences between the techniques used in coordinating MAS and those used in MRS are actually very few when the coordination mechanism is explicit (e.g. coordination protocols). So far, a number of review, taxonomy and survey analysis for MRS and MAS coordination has been published, some examples are the works of [4, 13] for classifying similarly MAS and MRS according to for instance communication, cooperativeness or decision autonomy. However, it is important to note that there are challenges and limitations to using MAS in robotics that must be carefully considered. One of the main challenges is that existing MAS coordination approaches may not be sufficiently adapted to handle uncertainty, acquire information from the environment, and model the incompleteness of robotics [16]. Therefore, it is important to identify a common framework that can effectively address these challenges. In addition, coordinating different types of robots can present an interesting challenge that requires careful consideration of multiagent coordination mechanisms. In this work, we develop the role of the agent embedding the learned action model in the multiagent part and present its place and use within the framework.

3 Hypothesis We consider a robot R and a factory F, with their physical and digital twins. The physical twin of R is noted R pt (resp. F pt ) and Rdt is its digital twin (resp. Fdt ). In the following, we use R or F to refer to both dimensions (see Fig. 1).

Fig. 1 Components and their connections according to our hypothesis

RoboTwin: Combining Digital Twin and Artificial Intelligence …

19

Fig. 2 Tools for the observation phase

Physical and digital dimensions are connected using robotic communication technologies (e.g. ROS1 based on the exchange of readable messages following a publish/subscribe protocol (Broker in Fig. 1). There is a complete connection loop between both dimensions: R pt publishes its status on topics to which Rdt subscribes; Rdt makes decisions according to this information and digital information in Fdt (DI in Fig. 1); Rdt publishes its decisions on topics to which R pt subscribes; R pt executes or not the decision according to the context in F pt . In this way, we consider the dynamics of the real environment. We know the goal of R and, considering it as a black-box, we do not know what is perceived by R pt and the Rdt action model. Thanks to a global observation system, we can collect information about its context and actions. Illustrative Example We consider the integration of a new AMR (R) in a factory (F) problem. According to our hypothesis, the features of our example are as follows: (1) The goal is the destination of R; (2) The status of R pt contains its location and direction, and the observed actions are its moves in F pt ; (3) Rdt decisions are the next robot movement; (4) DI concerns the digital map of F in Fdt .

4 Our Proposal We present a methodology and its tools to control the behavior of an autonomous robot in a factory. The methodology, the objectives of the tools and their connections are designed according to our hypothesis and the realization is done following our example. Our approach consists of three phases denoted as: (1) Observation (2) Learning (3) Integration. In the following sections, we explain each of these phases and the related tools.

1

The Robot Operating System (ROS), see https://www.ros.org.

20

F. Balbo et al.

4.1 Observation During the observation phase, we aim to generate logs that will be used in subsequent phases. We use three tools to manipulate scenarios (see Fig. 2): Builder. This tool is used to create a scenario, which sets the parameters for the Simulator and Logger. Simulator. This tool simulates the behavior of the robot R pt following the scenario. Rdt is connected to Simulator using the publish/subscribe protocol on the same topics as R pt . When Rdt makes a decision it publishes it. On reception, Simulator executes it, and publishes the new computed robot status with observations on the system (observation topic). Observations are related to all information that might influence Rd t decision. This last one being a black-box, we do not know what and how it is proceed but we know what are the input of the decision. Indeed to be included in the system, Rd t has subscribed to topics in the broker and has been connected to the factory information system. An observation is a collection of all the Rd t incoming information. Logger. This tool records data about the simulation by subscribing to the topics about R pt status (computed by the Simulator) and actions (computed by Rdt ) and obser vation. The result of this phase is the execution log, which is structured as a list of entries (CSV file format). Illustrative example. The specifications of the tools are: • Builder. Duration, robot locations (origin, destination), map definition, observation range are the simulation parameters. Observation range corresponds to the size of the discrete environment that is observed. This value may be different of the real size of the environment that is perceived by Rdt . In addition it supports the design of the related map (limits, obstacles (shape, location)). • Simulator. It updates the simulated robot location based on the movement published by Rdt and publishes the direction and location (status topic) and the content of the part of the environment that is in the observation range (obser vation topic). The simulation ends when R reaches its destination. • Logger. Figure 3 is a state of the system example that represents the 3r d row of Table 1. The x and y coordinates represent the location of the robot relative to the goal position, while orientation denotes the direction in which the robot is oriented. The neighborhood of the robot, where a cell (i, j) may have a numerical value vi, j defining the nature/identifier of an obstacle located at the position (i, j). If the cell is empty, it has a value of 0. Finally, decision denotes the action performed by the robot at time t.

RoboTwin: Combining Digital Twin and Artificial Intelligence …

21

Fig. 3 Context state example

Table 1 Execution log structure for integration of a new AMR problem example x

y

Orientation

(1,1)

...

(1,n)

(2,1)

…

(n,n)

Decision

84

17

up

0

...

0

0

...

−1

move_forward

84

16

up

0

...

0

0

...

0

turn_left

84

16

left

0

...

0

0

...

0

move_forward

...

...

...

...

...

...

...

...

...

...

0

0

right

1

...

0

0

...

0

stop

4.2 Learning During this phase, the R’s action model must be learned. To achieve this, we propose a tool called Learner, which aims to construct an action model by learning the pattern of taking an action for each specific context present in the log generated by the Logger. The choice of the learning model among learning models depends on the problem and is independent of the methodology. The output of this phase is a trained predictor (action model) that we define as a function that outputs an action (the predicted or most likely one) to be performed given a state (or a sequence of states). Illustrative example. The observable state of the system at time t is the input and we expect a movement as output. We have a limited, discrete output domain, which is the movement space. Therefore, the problem can be seen as a classification problem, where states are classified based on the movement they are expected to lead to. The logs of the robot can be seen as a labeled dataset to train algorithms to recognize specific patterns within the dataset and draw conclusions on how the states should be labeled or defined. Our problem can be seen as a supervised learning problem [3]. We can apply a feedforward machine learning algorithm, assuming that data points are independently and identically distributed. However, as an action performed at a step t may affect the state at every step t  > t, a data item can be dependent on those

22

F. Balbo et al.

Fig. 4 Tools for the learning and integration phases

that come before it. This type of information is also known as sequence data, and we therefore choose to apply LSTM. To construct the action model, we consider two spaces: the state space (location, orientation, neighborhood) and the movement space (decision) (see Table 1).

4.3 Integration In order to solve problems using the learned action model, it must be integrated into the loop between F pt and Fdt . The tool Decision, based on a MAS model, is responsible for anticipating potential problems and solving them through control of the robots. Decision subscribes to topics of the status of the robots and has the same level of information as Rdt , with access to the digital information in Fdt (Fig. 4). Each robot R is associated with an agent Ra that embeds the learned robot prediction model. Ra is able to anticipate the robot’s next states and make decisions accordingly. Decision for a problem may be used according to three automation levels: • without automation: Decision detects for the decision maker the future instances of the problem and supports the co-construction of the solution by computing the consequences of the modifications done on the system. The decision maker executes the solution. • semi-automation: Decision computes alternatives for solving the detected future problems, the decision maker chooses the best one and Decision executes it. • complete automation: Decision computes the best solutions to the detected future problems and executes it. The decision maker monitors the system. Illustrative example. Let us consider the without automation level. Decision presents to the decision maker the future locations of the agents computed using the learned action models. If a problem arises, the operator can add virtual obstacles on the map (Decision contains a copy of the Fdt map) to observe the modifications of the

RoboTwin: Combining Digital Twin and Artificial Intelligence …

23

Fig. 5 Example of indirect control

robot path. Once the decision maker is satisfied with the solution, he modifies the map in Fdt . Rdt will make decisions based on the new state of the map, thereby modifying the behavior of the robot. This manipulation allows the operator to fully control the execution scenarios by only acting on the digital environment without directly modifying the robot’s programming. The virtual obstacles are deleted when the conflicting initial situation ends.

5 Experiments To demonstrate the effectiveness of our methodology, we developed the tools of the proposed framework following the illustrative example described in the previous sections. Our framework is connected to the digital twin of a physical multi-robot system using TurtleBots.2 Both physical and digital systems use MQTT as a publish/subscribe communication protocol for data and action exchanges. We implemented the tools and the Rdt of one robot in Java. Its decision process is based on the shortest path computation with some randomness. The robot moves in the direction that decreases the distance to the destination and re-evaluate the decision to turn at random moments (not every moment). In Learner, we implemented an LSTM-based model. This network has a single hidden LSTM layer followed by a standard feed-forward network with probability outputs (one per class). The final output is the class applicable in the current state with the highest probability. This model achieved an accuracy of 78% on the validation set. Using these resulting predictors, we implemented an autonomous agent Ra in the tool Decision to manage the predicted action models (the GUI part of the tool is not yet implemented). Therefore, Ra can predict the robot action in advance and anticipate 2

https://www.turtlebot.com.

24

F. Balbo et al.

its future plans at runtime. For example, in Fig. 5a, we have a robot shown in blue, oriented down, and situated at (6,−3) relative to its goal shown in green. Based on its initial status and vision field, the agent in the tool Decision calls the action model to predict the next action and simulates performing this action to forecast the next state. Repetitively doing this, Ra obtains the sequence of the most likely actions to be performed until reaching the goal, which consists of 10 steps: moving forward (orientation is down) for 3 steps, turning right, making the orientation to the left, then continuing to move forward for the following 6 steps until reaching the goal. This information can be then used by the decision maker to indirectly control the robot movement. In this example, the decision maker succeeded in avoiding the incident area marked by the orange pentagon in Fig. 5b by forcing the robot to turn right earlier by placing the virtual gray obstacle in front of it.

6 Conclusions and Future Work In this work, we have presented RoboTwin, a new methodology and toolset for nonintrusively controlling the behavior of robots in Industry 4.0. Our proposal combines Digital Twin, Multi-Agent System, and Machine Learning models and techniques, and consists of three phases: Observation, Learning, and Integration. The proposed methodology can be adapted to several problems that require the control of robots. We applied ML methods to execution logs collected from digital environments, resulting in a behavior model that can be deployed for autonomous agent simulation, conflict resolution, optimization, and management purposes. We presented a proofof-concept implementation of this methodology and toolset for the Autonomous Mobile Robot scenario, testing it with the LSTM method and providing an example of the potential benefits of the resulting model. The framework we implemented provides the operator of a multi-robot system with tools for controlling the cyber-physical environment and anticipating the behaviors of autonomous entities. We believe that this methodology can be extended to various scenarios beyond robot navigation, using other advanced learning methods. In the short term, our focus is on improving the decision-making process in our proposal. Firstly, we will automate the control process by using a smarter agent that suggests the best positions for inserting virtual obstacles to avoid conflicts and optimize processes. Next, we will develop multi-agent solutions to improve the overall system when multiple agents are involved. Lastly, we plan to enhance the learning model to improve its efficiency.

RoboTwin: Combining Digital Twin and Artificial Intelligence …

25

References 1. Arora, A., Fiorino, H., Pellier, D., et al.: A review of learning planning action models. Knowl. Eng. Rev. 33 (2018) 2. Bolton, A., Butler, L., Dabson, I., Enzer, M., et al.: Gemini principles. CDBB_REP_006 (2018) 3. Cunningham, P., Cord, M., Delany, S.J.: Supervised Learning, pp. 21–49. Springer, Berlin (2008) 4. Farinelli, A., Iocchi, L., Nardi, D.: Multirobot systems: a classification focused on coordination. IEEE Trans. Syst. Man Cybern. Part B (Cybern.) 34(5), 2015–2028 (2004) 5. Fragapane, G., Ivanov, D., Peron, M., et al.: Increasing flexibility and productivity in industry 4.0 production networks with autonomous mobile robots and smart intralogistics. Ann. Oper. Res. 1–19 (2020) 6. Fuller, A., Fan, Z., Day, C., Barlow, C.: Digital twin: enabling technologies, challenges and open research. IEEE Access 8, 108952–108971 (2020) 7. García-Martínez, R., Borrajo, D.: An integrated approach of learning, planning, and execution. J. Intell. Robot. Syst. 29(1), 47–78 (2000) 8. Gerkey, B., Mataric, M.J.: Are (explicit) multi-robot coordination and multi-agent coordination really so different. In: Proceedings of the AAAI Spring Symposium on Bridging the Multi-agent and Multi-Robotic Research Gap, pp. 1–3 (2004) 9. Ma, Y., Zhu, X., Zhang, S., Yang, R., et al.: TrafficPredict: trajectory prediction for heterogeneous traffic-agents. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, pp. 6120–6127 (2019) 10. Manh, H., Alaghband, G.: Scene-LSTM: a model for human trajectory prediction. arXiv:1808.04018 [cs] (2019) 11. Medsker, L.R., Jain, L.: Recurrent neural networks. Des. Appl. 5, 64–67 (2001) 12. Mourao, K., Petrick, R.P., Steedman, M.: Using kernel perceptrons to learn action effects for planning. In: International Conference on Cognitive Systems (CogSys 2008), pp. 45–50. CiteSeer (2008) 13. Ota, J.: Multi-agent robot systems as distributed autonomous systems. Adv. Eng. Inf. 20(1), 59–70 (2006) 14. Safaei, J., Ghassem-Sani, G.: Incremental learning of planning operators in stochastic domains. In: International Conference on Current Trends in Theory and Practice of Computer Science, pp. 644–655. Springer, Berlin (2007) 15. Tran, H., Le, V., Tran, T.: Goal-driven long-term trajectory prediction. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 796–805 (2021) 16. Verma, J.K., Ranga, V.: Multi-robot coordination analysis, taxonomy, challenges and future scope. J. Intell. Rob. Syst. 102(1), 1–36 (2021)

A Method of Positioning a Humanoid Robot Relative to the Center of a Group of People—Analysis and Implementation Marko Kovaˇcevi´c and Zdenko Kovaˇci´c

Abstract Here we describe the mathematical model of the humanoid robot positioning strategy guided by the idea of adopting a new position with respect to the center of a group of people. We show that such a type of positioning changes for different numbers of people, where the cases studied include one, two, three, and four or more people in a group. We conclude the paper with the implementation of the group center positioning method on the humanoid robot Pepper and the tests under laboratory conditions showed the naturalness and effectiveness of the chosen positioning strategy.

1 Introduction The success of robot-human interaction depends on where it takes place, the direction in which humans and robots move, and their speed, as well as the context in which the interaction takes place. The aim of research described in this paper was to deepen the knowledge of successful positioning relatively to a group of persons with whom the robot wants to interact. Approach strategies have been the subject of research by many research groups [1– 9], and when robots are considered, this is all the more important because some robots, being large and heavy, can pose a real physical threat to humans [10]. For example, home assistance robots are by definition supposed to be harmless to humans, and under such circumstances, approach strategies have a significance that goes beyond the ergonomic and safety aspects [11–20]. The way the robot navigates in the vicinity of humans depends on the role of the robot, more precisely on the type of interaction the robot tries to establish [21]. Robots M. Kovaˇcevi´c (B) · Z. Kovaˇci´c Faculty of Electrical Engineering and Computing, University of Zagreb, 10000 Zagreb, Croatia e-mail: [email protected] URL: https://www.fer.unizg.hr/zdenko.kovacic Z. Kovaˇci´c e-mail: [email protected] URL: https://www.fer.unizg.hr/zdenko.kovacic © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_3

27

28

M. Kovaˇcevi´c and Z. Kovaˇci´c

can interact with humans as their companions, talking to them, instructing or guiding them while navigating in closed or open space [22–24]. From the human point of view, the robot’s navigation behaviour must look “natural”. The movements must be soft and adaptable, in a word “human-like” [25, 26]. Just as humans learn how to communicate/interact with other humans from an early age, the robot can do the same. In [27, 28] learning techniques such as reinforcement learning are used to achieve socially adaptive path planning for approaching humans. Adaptation of robot behaviour has become an important research topic in the field of human-robot interaction [29, 30]. The problem of approaching a group of people involved in a discussion or other type of interaction was treated in [31] as part of a mission consisting of entering a room, approaching the people, being in the group and leaving the group, and continuing to navigate the environment. The described approach technique aligns the robot’s direction of travel with the direction of the detected person. In case of approach of more than one person, the robot has been guided so that it faces the center of the group. This was achieved by defining the direction of travel of the robot by averaging the directions of travel of each individual person.

1.1 Contribution Although it seems that the positioning of the robot next to a group of humans during establishing contact is a well studied problem, we have found that it can still be addressed in a deeper elaborated analytical way. A chosen robot positioning strategy implies that the robot positions itself at the edge of the best-fitting circle of the group (with the center in a mean circle center point). Therefore, the main contribution of this paper is the mathematical formulation of this strategy that allows implementation on a real humanoid robot in such a way that the robot can naturally and comfortably initiate interaction with individuals or groups of humans. The mathematical study also considers proxemics theory to ensure that the robot stays within the social distance of the people: not so close as to be uncomfortable, and not so far as to impede communication. The recorded experiments compiled in the video supplement show the cases of effective robot positioning for a group of one/two/three/four people implemented on the humanoid robot Pepper within the Robot Operating System (ROS) environment.

1.2 Organization of the Paper The paper is structured as follows. First, we analyze the requirements that a robot must satisfy in order to position itself near humans. Then, we define the mathematical model of a chosen relative positioning strategy. Furthermore, we describe experimental results obtained on the humanoid robot Pepper. We conclude the paper with comments on the obtained results and present some ideas for future work.

A Method of Positioning a Humanoid Robot Relative to the Center …

29

2 Problem Definition We address the problem of relative positioning the robot with respect to the group of people {H1 , . . . , Hn }, while allowing other robots {R1 , . . . , Rn } to be there. Humans and robots are in open unconstrained two-dimensional space. The positions of the people are expressed in the local coordinate system of the robot in the form Ri [x HRij , y HRij , ϕ H j ], j = 1, . . . , n. Similarly, the position of the robot Ri is expressed Ri Ri as [x Ri , y Ri , ϕ Ri Ri ], i ∈ [1, m]. We assume that the position of the robot in space is determined. The following assumptions must be considered: • A group of people is standing in position and slightly moving their feet. An example might be a reception with drinks and snacks, where people stand in groups of somewhat circular shape or in the arc formation facing the center of the room. People interact with each other, talk to each other, move their heads, and interact with other members of the group. • The robot has access to accurate data [x HRij , y HRij , ϕ H j ] about the positions and directions of the human heads. All coordinates are in the local robot system whose origin is between the legs of the robot. • We consider here the case of a single robot that is positioning itself with respect to a human, i.e., m = 1. • The robot has unobstructed access to the group, i.e., the space consists only of the people in the group and the robot. • When interacting with humans, robots must respect people’s personal space and be generally mindful of their comfort [7]. Nowadays, perhaps the most acceptable approach is to respect the four proxemic zones [2, 3] shown in Table 1. • We assume that the persons and the robot have not met and are not yet familiar with each other up to this point. The method described in the paper determines the distance between robots and people based on the formation in which the people are standing. The robot mimics the human in the sense that it positions itself as close as possible to the group members, and the upper boundary of the personal zone is taken as the minimum distance. The designed method would have a wider scope if it allowed an additional degree of freedom, namely the distance to the persons. Therefore, the desired distance is implemented as parameter d ∗ .

Table 1 Proxemic zones defined by Hall [2] Zone Distance Intimate Personal Social Public

0–0.45 m 0.45–1.2 m 1.2–3.6 m > 3.6 m

Situation Love and close friendship Conversation between friends Conversation Public speaking

30

M. Kovaˇcevi´c and Z. Kovaˇci´c

2.1 Robot Positioning Relative to a Single Person In the case of a single person approached by the robot, the simplest solution is for the robot to come before the person. The method is based on a simple principle where the desired distance to the person and the angle relative to the person’s orientation are used as polar coordinates in a system whose origin is between the person’s legs. Finally, the resulting reference coordinates x ∗ , y ∗ , and ϕ∗ are transferred to the local robot system as follows: ∗R1 H1 R1 = x HR11 + d ∗ cos(ϕapp + ϕH x R1 1)

(1)

∗R1 H1 R1 y R1 = y HR11 + d ∗ sin(ϕapp + ϕH 1 ))

(2)

H1 R1 1 ϕ∗R R1 = ϕapp + ϕ H 1 + π

(3)

It is important to note that these computed coordinates are only valid until the robot moves, as this changes the origin of the robot’s local system, requiring a new computation. In practical implementation, this is not a problem, as the robot’s builtin program takes this into account. Under the above assumptions, there is no case where this method would not give a satisfactory result.

2.2 Robot Positioning Relative to Two Persons Research in [33] has confirmed that couples do not feel comfortable when the robot approaches from behind. When two people are talking, we assume that most likely they will be looking at each other or at something of common interest in the environment. Based on this assumption, a relative positioning method that we want to analyze will use the head positions of both persons to determine on which side of the straight-line connecting the persons’ the robot should go. If the robot does not know the persons beforehand and has no preference for a person, the robot will position itself at the same distance from both persons. This implies that the persons and the robot will form a triangle. Considering that the distance of the robot affects the person [7], it is implied that there should be a control over this parameter. For example, forcing a robot to move and become the corner of an equilateral triangle does not allow for such control, as the distance from robots to persons is closely related to the distance between humans. In general, the distance between two persons can be much closer than they would like to have a robot near them. Therefore, we have chosen the isosceles triangle approach, where the position of the third corner of the triangle on which the robot stands is calculated taking into account the proxemic zones relative to the two persons H1 and H2 , whose positions correspond to the positions of the other two corners of the triangle. The data about these positions are provided by a robot R1 whose goal is to interact with the

A Method of Positioning a Humanoid Robot Relative to the Center …

31

individuals. To position the robot in front of the subjects, it is necessary to determine the angle β of the line on which the subjects are standing and the symmetry of the R1 R1 angle γ closed by the view angles of the subjects ϕ H 1 and ϕ H 2 . β = a tan 2(y HR11 − y HR12 , x HR11 − x HR12 )  γ=

R1 | δ − ϕH 1 |≤ R1 | δ − ϕ H 1 |>

π 2 π 2

: δ : δ+π

δ=

R1 R1 ϕH 1 − ϕH 2 2

(4) (5)

By comparing the obtained angles, the orientation of the robot is determined such that the robot body is aligned with the center of the line segment H1R1 H2R1 . Finally, the robot is set to distance d from the center point at angle ϕ∗R1 R1 + π. Based on the head orientation of both persons, it can be determined on which side of this line the robot should position itself, and considering the lower bound dmin of the robot’s distance to the person, the robot position is calculated as follows: α=β−γ ϕ∗R1 R1 =



(6)

α≤π:β− α>π:β+

π 2 π 2

(7)

 d=

d H 1,H 2 ≥ dmin : d H 1,H 2 d H 1,H 2 < dmin : dmin

d H 1,H 2 =| H1R1 H2R1 |

(8)

∗R1 = x R1

x HR11 + x HR12 + d cos(ϕ∗R1 R1 + π) 2

(9)

∗R1 = y R1

y HR11 + y HR12 + d sin(ϕ∗R1 R1 + π) 2

(10)

Figure 1 compares where people are located within the different proxemic zones. It is obvious that the isosceles triangle method respects people’s personal zones. Following the reasoning in [7], to obtain an influence on the positioning of robots with respect to a pair of persons, instead of using Eq. (7), the angle ϕ∗R1 R1 can be calculated in the following way: ϕ∗R1 R1 =

 H1 α π : β + 2 + ϕapp

H1 The angle ϕapp is chosen after [33] or one can choose another way.

(11)

32

M. Kovaˇcevi´c and Z. Kovaˇci´c

Fig. 1 Test examples using the isosceles triangle method of robot positioning. Left: an example where a robot is positioned at the distance of the upper boundary of the personal zone, even though the people are at the boundary of the intimate zone. Right: when the people are far enough away, the isosceles triangle method is similar to the equilateral triangle method

2.3 Positioning Next to Three or More Persons When the group consists of three or more people, the situation is more complex and varied. Unlike two persons who can only stand at the ends of the line segment, three people can stand in a line or in triangles of various kinds, while four or more people can stand in even more diverse formations. In this section, we discuss a methodthat  computes the default position of a robot R1R1 based on known H1R1 , . . . , HnR1 for n ≥ 3. Robot Positioning Relative to the Center of a Group of People The approach to the relative positioning of a robot with respect to a group of people is based on the idea that the positions of any three people lie on a common circle. If someone, in this case a robot, wants to interact, it is implied that it wants to become the part of a group on this circle. Based on this consideration and the fact that the circle is uniquely defined by only three points, it makes sense for the robot to position itself with respect to the center of this group (circle center). Since it is not possible to uniquely determine the circle by four points because of the need to solve a four-equation system with three unknowns, the case of n people in the group is further divided into the subcases n = 3 and n ≥ 4. In the case of a group of three persons, their positions are sufficient to find the center of the group by finding the center (xc , yc ) and the radius r of the parametrically defined circle: (12) x 2 + y2 + A ∗ x + B ∗ y + C = 0 where xc =

 A B , yc = , r = xc2 + yc2 − C 2 2

(13)

The determination of the robot’s position takes into account the radius r of the common circle and the angles δi that the subjects take with respect to the center. The goal of the calculation is for the robot to “close” the largest distance between

A Method of Positioning a Humanoid Robot Relative to the Center …

33

the people and thus become part of the formation. Specifically, the method finds the largest angle δmax between two adjacent persons with respect to the center of circle: δimax = max {δ1 , . . . , δn }

(14)

To calculate these angles δi , the group of people HiR1 is sorted in ascending order R1 by the angle of view ϕ H i , and the angle δi is obtained as δi =| Φi − Φ(i+1+n) |

i ∈ {1, . . . , n}

Φi = atan2(y HR1i − yc , x HR1i − xc )

(15) (16)

Using (13), relative to the centerline of the angle δmax β = Φimax +

δimax 2

(17)

the robot positions itself as follows ∗R1 x R1 = xc + r cos(β)

(18)

∗R1 = yc + r sin(β) y R1

(19)

The angles that the individuals take with respect to the center of the circle are used to compute the position of the robot on the circle such that the robot "closes" the largest distance between the individuals and becomes part of the formation. In reality, when positioning near a group of four or more people, one cannot expect all of their positions to be on the circle, and with even more people, the probability is even n  lower. For this reason, the circle center method for n = 3 is extended such that combinations of trios of persons determine the corresponding circles. The center 3 of such a group is obtained as the mean of all circle centers computed with (12) and (13): n n n 1    Tc,i jk T¯c = (x¯c , y¯c ) = n  3

(20)

i=1 j=i+1 k= j+1

where Tc,i jk is the center of the circle determined by the positions of the trio of persons HiR1 , H jR1 , and HkR1 . The mean distance r¯ is defined as the arithmetic mean of all distances of the individuals from the group center T¯c : r¯ =

n  1 )2 + ( y¯c − y H(R1) )2 (x¯c − x H(R1) i i n i=1

(21)

The rest of the method is the same as for the circle center method for n = 3.

34

M. Kovaˇcevi´c and Z. Kovaˇci´c

Fig. 2 Test examples of evaluating a circle center method for positioning with respect to three people, shown in the robot’s local coordinate system. Left: the case of a closed formation of individuals. Right: the case of an open formation of people. The blue symbol shows the calculated position of the robot R1∗R1 , the green shows the current position of the robot, and the black symbols show the positions of the people HiR1 . The red circle represents the computed circle, while the cross indicates its center

The mean circle center of the group method was also tested in the simulation environment on instances that resemble real-world situations. We again changed the position of the group of persons in relation to the robot and then selected and analyzed edge examples that were expected to lead to different solutions. The left graph in Fig. 2 shows that the method satisfies very well when people are in a closed formation and facing the center of the group. In the righthand graph, the individuals are in an open formation and it can be seen that the method gives the target position of the robot too far away from the group. In such a case, the solution is to constrain the value of d ∗ (force the robot closer to the center of the circle) while taking care to stay within the social zone. Figure 3 shows the results of positioning near a group of four and five persons standing in a closed formation (left and middle graphs). It can be seen that the mean circle center method satisfies very well in both cases. In turn, with an open formation of four people, the target position of the robot is determined too far away from the group (see right graph). The solution to this problem is again to limit d ∗ and force the robot to approach the center of the circle until it reaches the boundary of a social zone. Further investigation of the circle center method has found one case where a bifurcation occurs, i.e., the method has a discontinuity here. In two similar cases shown in Fig. 4, when the position of one person (the second person on the left) changed a little, the position of the robot changed a lot. The reason for this is that when three people are almost colinear, a small displacement of one person can cause the arc to move from concavity to convexity and vice versa. This results in a very large displacement of the center of the circle, which directly affects the position of the robot. A possible solution to this problem, in order to increase robustness, is to limit the amount of radius calculated in (21).

A Method of Positioning a Humanoid Robot Relative to the Center …

35

Fig. 3 Test examples of evaluating a mean circle center method for positioning with respect to n ≥ 4 people, shown in the robot’s local coordinate system. Left: the case of a closed formation of four people. Middle: the case of a closed formation of five people (without circles drawn in for better visibility). Right: the case of the open formation of four people. The blue symbol shows the calculated robot position R1∗R1 , the green one shows the current position of the robot, and the black symbols show the positions of the people HiR1

Fig. 4 Two similar test examples, plotted in the robot’s local coordinate system, showing the discontinuity of the circle center method. In the bottom graph, you can see that the second person on the left has slightly changed position. The blue symbol shows the calculated robot position R1∗R1 , the green shows the current position of the robot, and the black symbols show the positions of HiR1

3 Laboratory Experiments The laboratory experiments took place in the premises where the lighting conditions are similar to those expected in any indoor space intended for social contact. The main goal of the experiments was to verify whether the mathematically formulated method of positioning according to the best fitting circle of the group could be successfully implemented on one of the more advanced humanoid robots. During the experiments, the volunteers were asked to freely create a group in which they felt most comfortable. For practical reasons, in order to know the exact starting location

36

M. Kovaˇcevi´c and Z. Kovaˇci´c

of the robot, the robot always moved from the same starting position and, using its sensors, first detected the people and their positions and then calculated the position to reach. Individual situations, especially when there were three and four people in the group, were repeated several times and the observed problems were solved by modifying the parameters and boundary conditions when calculating the robot’s position. In experiments, it was found that too frequent position adjustments of the robot, triggered by wiggling of the people, eventually led to wrong robot positioning. Thorough research has shown that the reason for this is that the robot cannot observe the people while they are moving, and therefore, in the absence of sufficient time for recalculation, it does not update the people’s positions correctly. To prevent the robot from positioning itself based on the old and inaccurate positions of the people, the implemented method checks if the current position of the people is the same as when the method was last called. If this is the case, it means that the positions of the people have not been updated and accordingly the method does not place the robot at a new position. There is a video supplement to this post that contains the recording of the experiments [34]. After all upgrades, the robot positioned itself according to the idea. The volunteers who participated in the test felt comfortable in Pepper’s company. The problem was Pepper’s narrow field of vision. The volunteers first had to stand in front of Pepper or ask him to observe them and put them on the list of people (see Fig. 5). After that, they could slowly move around the room and Pepper would continue to observe them. When calling the implemented method handleGoT oGr oup, the robot would begin to move toward the group. Figure 6 shows some snapshots from the video supplement showing experiments conducted in the presence of one, two, three and four people.

Fig. 5 Pepper makes eye contacts with group members to memorize them

A Method of Positioning a Humanoid Robot Relative to the Center …

37

Fig. 6 Relative positioning of a Pepper robot in the presence of one, two, three and four persons— snapshots from the video supplement [34]

4 Conclusion In the field of robot-human interaction, there are a number of properties that a robot must have in order to achieve a successful interaction. This requires sufficient physical proximity to humans and awareness of the presence of humans in its vicinity. Physical proximity is achieved by positioning near humans, and the way in which a robot performs this is very important. In this work, we have performed the analysis of a strategy for positioning near the people with respect to the mean circle center of the group. For this strategy, we derived a mathematical model in a systematic way, based on the typical sensory apparatus of contemporary robots. We then used this mathematical model to validate the method in the simulation environment. We found that the mean circle center method is robust enough to be implemented on the Pepper humanoid robot. Although the Pepper robots have many advanced features that aid in the implementation of the positioning strategy presented in this paper, there is clearly much room for further improvement and development of new features. In particular, this highlights the need for a larger field of view and faster image processing and interpretation of dynamic changes due to unpredictable movements of people, occlusions, position changes, and much more. We plan to continue our work on improving localization and path planning algorithms. We also plan to improve the cognitive capabilities of the Pepper robots using state-of-the-art artificial intelligence and machine learning techniques.

38

M. Kovaˇcevi´c and Z. Kovaˇci´c

Nonetheless, our research has ended with a successful proof of concept. We have shown that a robot like Pepper, designed to interact with humans, can position itself near the humans in a very natural way, fulfilling the first requirement for contact.

References 1. Walters, M.L., Dautenhahn, K., Woods, S.N., Koay, K.L.: Robotic etiquette: results from user studies involving a fetch and carry task. In: HRI’07, Arlington, Virginia, USA, pp. 317–324, 8–11 Mar 2007 2. Hall, E.T.: The Hidden Dimension: Man’s Use of Space in Public and Private. The Bodley Head Ltd., London, UK (1966) 3. Hall, E.T.: Proxemics. Curr. Anthropol. 9(2–3), 83–108 (1968) 4. Gillespie, D.L., Leffler, A.: Theories of non-verbal behavior: a critical review of proxemics research. Socialogical Theory 1, 120–154 (1983) 5. Hediger, H.: The evolution of territorial behavior. In: Washburn, S.L. (ed.) Social Life of Early Man, pp. 34–57. Butterworth and Co., London (1961) 6. Burgoon, J.K., Jones, S.B.: Toward a theory of personal space expectations and their violations. Hum. Commun. Res. 2(2), 131–146 (1976) 7. Walters, M.L., Koay, K.L., Woods, S.N., Syrdal, D.S., Dautenhahn, K.: Robot to human approaches: preliminary results on comfortable distances and preferences. In: Multidisciplinary Collaboration for Socially Assistive Robotics (2007) 8. Satake, S., Kanda, T., Glas, D.F., Imai, M., Ishiguro, H., Hagita, N.: How to approach humans? Strategies for social robots to initiate interaction. In: Proceedings of the 4th International ACM/IEEE International Conference on Human-Robot Interaction HRI’09, La Jolla, California, USA, pp. 109–116, 11–13 Mar 2009 9. Tanmay, R., Bera, A., Manocha, D.: F2fcrowds: planning agent movements to enable face-toface interactions. Presence Teleoperators Virtual Environ. 26(2), 228–246 (2017) 10. Nomura, T., Shintani, T., Fujii, K., Hokabe, K.: Experimental investigations of relationships between anxiety, negative attitudes, and allowable distance of robots (HCI 07). In: Proceedings of the Second International Conference on Human Computer Interaction (HCI’07), Beijing, P.R. China, pp. 13–18 (2007) 11. Jeppesen, K., Bodenhagen, L., Krãijge, N.: Socially acceptable behaviour for robots approaching humans using an adaptable personal space. In: Proceedings of the 12th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP’2017), pp. 116–121 (2017) 12. Mumm, J., Mutlu, B.: Human-robot proxemics: physical and psychological distancing in human-robot interaction. In: 6th International Conference on Human-Robot Interaction (2011) 13. Mead, R., Mataric, M.J.: Autonomous human-robot proxemics: socially aware navigation based on interaction potential. Auton. Robots 41, 1189–1201 (2017) 14. Hüttenrauch, H., Severinson Eklundh, H., Green, A., Topp, E.A.: Investigating spatial relationships in human-robot interaction. In: Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS’06), Bejing, China, pp. 5052–5059 (2006) 15. Michalowski, M.P., Sabanovic, S., Simmons, R.: A spatial model of engagement for a social robot. IEEE International Workshop on Advanced Motion Control (AMC’06), pp. 762–767 (2006) 16. Tasaki, T., Matsumoto, S., Ohba, H., Toda, M., Komatani, K., Ogata, T., Okuno, H.G.: Dynamic communication of humanoid robot with multiple people based on interaction distance. IEEE International Workshop on Robot and Human Interactive Communication (Ro-Man 2004), pp. 71–76 (2004)

A Method of Positioning a Humanoid Robot Relative to the Center …

39

17. Gockley, H.G., Forlizzi, J., Simmons, R.: Natural person following behavior for social robots. In: Proceedings of the International Conference on Human-Robot Interaction (HRI2007), pp. 17–24 (2007) 18. Sisbot, E.A., et al.: Navigation in the presence of humans. In: IEEE-RAS International Conference on Humanoid Robots (2005) 19. Henry, P., Vollmer, C., Ferris, B., Fox, D.: Learning to navigate through crowded environments. In: 2010 IEEE International Conference on Robotics and Automation (ICRA), pp. 981–986, May 2010 20. Rios-Martinez, J., Spalanzani, A., Laugier, C.: From proxemics theory to socially-aware navigation: a survey. Int. J. Soc. Robot. 7(2), 137–153 (2015) 21. Kruse, T., Pandey, A.K., Alami, R., Kirsch, A.: Human-aware robot navigation: a survey. Rob. Auton. Syst. 61(12), 1726–1743 (2013) 22. Repiso, E., Ferrer, G., Sanfeliu, A.: On-line adaptive side-by-side human robot companion in dynamic urban environments. In: IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 872–877 (2017) 23. Kessler, J., Schroeter, C., Gross, H.M.: Approaching a person in a socially acceptable manner using a fast marching planner. In: Jeschke, S., Liu, H., Schilberg, D. (eds.) Intelligent Robotics and Applications, ICIRA’2011. Lecture Notes in Computer Science, vol. 7102 (2011) 24. Ferrer, G., Garrell, A., Herrero, F., Sanfeliu, A.: Robot social aware navigation framework to accompany people walking side-by-side. Auton. Rob. 1–19 (2016) 25. Ramon-Vigo, R., Perez-Higueras, N., Caballero, F., Merino, L.: Transferring human navigation behaviors into a robot local planner. In: 2014 RO-MAN: The 23rd IEEE International Symposium on Robot and Human Interactive Communication, pp. 774–779 (2014) 26. Bennewitz, M., Burgard, W., Cielniak, G., Thrun, S.: Learning motion patterns of people for compliant robot motion. Int. J. Rob. Res. 24(1), 31–48 (2005) 27. Kim, B., Pineau, J.: Socially adaptive path planning in human environments using inverse reinforcement learning. Int. J. Soc. Rob. 1–16 (2015) 28. Ramírez, O.I., Khambhaita, H., Chatila, R., Chetouani, M., Alami, R.: Robots learning how and where to approach people. In: 25th IEEE International Symposium on Robot and Human Interactive Communication (RO-MAN), pp. 347–353, Aug 2016 29. Svenstrup, M., Tranberg, S., Andersen, H.J., Bak, T.: Pose estimation and adaptive robot behaviour for human-robot interaction. In: IEEE International Conference on Robotics and Automation, pp. 3571–3576 (2009) 30. Narayanan, V.K., Spalanzani, A., Luo, R.C., Babel, M.: Analysis of an adaptive strategy for equitably approaching and joining human interactions. In: 2016 RO-MAN: The 25th IEEE International Symposium on Robot and Human Interactive Communication, pp. 341–346 (2016) 31. Althaus, P., Ishiguro, H., Kanda, T., Miyashita, T., Christensen, H.: Navigation for human-robot interaction tasks. In: IEEE International Conference on Robotics and Automation, vol. 2, pp. 1894–1900 (2004) 32. Shao, X., et al.: Detection and tracking of multiple pedestrians by using laser range scanners. In: International Conference on Intelligent Robots and Systems (IROS’2007), pp. 2174–2179 (2007) 33. Joosse, M., Lohse, M., van Dijk, B., Evers, V., Karreman, D., Utama, L.: Robot, etiquette: how to approach a pair of people? In: 9th ACM/IEEE International Conference on Human-Robot Interaction, HRI’2014, pp. 196–197 (2014) 34. Video supplement: humanoid robot positioning next to a group of 1/2/3/4 people based on the mean circle center method. https://www.youtube.com/watch?v=4oJkc5Xx-dQ. Posted in Apr 2021

How Can Adults Make Time to Study: A System for Employee Sharing and Reskilling Education Kenta Abe , Minoru Matsui , and Hisashi Hayashi

Abstract With significant developments in AI and machines, workers increasingly need to continuously study and learn. However, studying in one’s spare time while working can often be challenging. Employees could make time for study by taking extended paid leave to attend school, but companies typically find it difficult to fill vacancies in such cases. In this study, we propose an employee exchange system based on an associated labor token to balance these issues. In this system, companies receive tokens from the government based on the number of employees who have taken leave. The company then uses those tokens to borrow employees from other companies participating in this system for a month. The company that rents the employee receives the tokens and can use them in the same way. The effectiveness of this approach was verified by simulation, and the results showed that employees were effectively given enough time to complete their studies at school. Companies were also able to effectively conduct business by exchanging employees in this manner. Furthermore, the results suggested that the financial costs were lower than hiring new people. Future research should include simulations with different countries, regions, and types of jobs.

K. Abe (B) · M. Matsui · H. Hayashi Advanced Institute of Industrial Technology, 1-10-40 Higashi-Ooi, Shinagawa-Ku, Tokyo 140-0011, Japan e-mail: [email protected] M. Matsui e-mail: [email protected] H. Hayashi e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_4

41

42

K. Abe et al.

1 Introduction With the rapid development of AI and machines, various human jobs are being automated. Various studies have been conducted on whether AI and machines will replace human jobs, with both pessimistic and optimistic predictions [1–9]. In any case, working people now need to continue studying more than ever. Several issues must be considered when working people attend school. Employees may find it difficult to make time to study while working; for employers, filling vacancies and managing costs can be challenging. Taking Japan as an example, the Ministry of Health, Labor and Welfare has established an “education and training leave system” [10]. Under certain conditions, the government provides subsidies to companies that provide their employees paid leave for education and training. This program is noteworthy for employees who wish to acquire new skills. However, few companies have adopted the program [11], primarily owing to the difficulty in securing replacement personnel [12]. In this study, we propose an intercompany employee exchange method as a sociotechnological system to solve these problems and examined its effectiveness using simulations. Specifically, (1) employees in occupations highly substitutable for assistance by AI or machines are allowed to take long-term paid leave to attend school, and (2) a company with vacancies caused by long-term paid leave temporarily replaces members of its workforce through an intercompany employee exchange. For (1), we verified the effectiveness of this approach by comparing the number of students who complete their studies with and without introducing the company employee exchange program. Regarding (2), we examine the effectiveness of the employee replacement program by comparing the number of employees who completed the study period with the number of employees who did not. Several researchers have argued that AI, machines, and other technologies may replace human workers, even though their estimates differ. While some studies predict that a certain number of replacements will occur [1–5], others predict that replacements will occur owing to factors other than the introduction of AI and machines [6, 7]. Certain studies have also pointed to the importance of reshaping the competencies of employees who may be replaced by AI [8], while others have noted that AI will allow them to focus on strategic tasks [9]. In other words, regardless of whether AI and machines are implemented to perform human jobs, humans must continue to learn. Many previous studies have been conducted on learning and associated difficulties for working people [13–16] as well as on staffing [17–25], in particular, [19– 25] discuss proposed algorithms for staffing, while [21–25] validated the developed algorithms in a software or a simulator. However, none of these studies discuss how working adults can find time to study or how to fill in for employees who have taken time off to attend school. In addition, as pointed out in [23–26], tasks and resources can be distributed in either a centralized or decentralized fashion. In the centralized approach, a control center is responsible for task and resource distribution for each agent. In contrast,

How Can Adults Make Time to Study: A System for Employee Sharing …

43

in the decentralized approach, each agent acts cooperatively or independently to achieve an optimal task and resource distribution. In this study, we adopt a centralized approach because the proposed system is assumed to be a government project, in which the government coordinates the requirements of the various companies. Furthermore, as this project involves government subsidies, it is more realistic to assume that would have to be under government control. The remainder of this study is organized as follows. In Sect. 2, we explain the proposed employee exchange system using virtual tokens to allow for educational leave. In Sect. 3, we describe the configuration of the experimental simulation. In Sect. 4, we detail the method used to match companies that need to borrow employees and those that can lend them. In Sect. 5, completion of an educational leave is defined. In Sect. 6, the results of the experiment are discussed. In Sect. 7, we provide some concluding remarks and suggest some possible directions for future research.

2 Token/Employee Exchange System to Allow Study Time In this study, we propose a method to fill in for employees who take full-year paid leave to attend school. Specifically, we propose an intercompany employee exchange system based on virtual token exchange. We evaluated the effectiveness of the proposed approach in a simulation. This system is inspired by the work of Saito et al. [23], which assumes that the number of employees of all participating companies does not increase or decrease. We assume a certain number of vacancies due to employees attending school each year.

2.1 System Overview The purpose of this system is to provide several employees with paid time off for attending school each year. Furthermore, this system is designed to ensure the availability of labor through an intercompany employee exchange. The government functions as the administrator of the system and manages it on an annual basis. Each company provides paid study leave to all employees (candidate workers) whose jobs may be replaced by AI or machines in the near future. Upon completion of the study, employees return to their original company and begin working again, which means that each company participating would have several vacancies each year. To fill these vacancies, companies send employees to each other during busy and slow periods.

44

K. Abe et al.

Fig. 1 Explanatory diagram of token/employee exchange system

2.2 Intercompany Employee Exchange Using Virtual Tokens The government issues a certain number of virtual tokens to companies participating in this project once at the beginning of each fiscal year. Each company has 10–20% candidate workers of all workers (randomly set by uniform distribution), and the number of tokens depends on the number of paid vacations each company provides to a certain number of candidate workers for one fiscal year (Fig. 1a). These tokens are returned to the government in exchange for a given number of workers when a company is short of staff and asks to borrow people from other companies (Fig. 1b). The government matches companies that have requested personnel (borrowing companies) with companies that have responded that they are willing to lend people (lending companies) for reasons such as off-peak periods. If the match is successful, the lending company receives tokens from the government and lends the borrowing company a number of employees equal to the number of tokens on a month-to-month basis (Fig. 1c). Each company can borrow or lend employees depending on business needs, which means that the distribution of tokens will occur. The overall number of tokens does not increase or decrease within a fiscal year. Tokens reset yearly for two reasons. First, employees can only take leave on an annual basis. In a new fiscal year, several other candidate workers take leave. Therefore, the previous year’s tokens must be reset, and a new set of tokens issued for the new year. Second, government budgets operate on an annual basis. Therefore, using a standard fiscal year cycle to manage the issuance and resetting of tokens is the most practical approach.

2.3 Workers Who Attend School In the present study, we refer to clerical and other employees without specific professional skills as candidate workers, referring to [4]. All employees, including candidate workers, have the following attributes (Table 1).

How Can Adults Make Time to Study: A System for Employee Sharing …

45

Table 1 Attributes of each worker Attribute

Definition

Free time

24 h, distributed as follows Working: 8 h Allowable overtime: randomly set by normal distribution with a mean of 1 h, standard deviation of 0.5 h, minimum of 0 h, and maximum of 3 h Commute: randomly set by uniform distribution from 0.5 to 1.5 h Meals and housework: randomly set by uniform distribution from 2.5 to 3.5 h Sleep: randomly set by uniform distribution from 6.5 to 7.5 h

Motivation value

A hypothetical quantification of person’s motivation to attend school. This is randomly set by normal distribution with a mean of 50, a standard deviation of 10, a minimum 0, and maximum 100

When the motivation value exceeds the threshold, each employee tries to cut their overtime in half and devote their remaining free time to study. Assume that school requires 1 h of preparation and 1.5 h per class; therefore at least 2.5 h per day is necessary. Candidate workers are exempt from working for a year, so they study five classes per day (8.5 h total). Normal workers who are not candidate workers take as many classes as possible when they have at least 2.5 h of free time per day available. On Saturday, both candidate workers and normal workers whose motivation value is over 70 study at school up to 8.5 h. On Sunday, both workers don’t study at school.

2.4 Salary Paid by the Company In the simulation, we considered a salary of $20.00 per hour, with overtime pay multiplied by 1.2. In this simulation, this hourly rate is company-wide. Employees are paid a salary as follows. An employee lent to another company has a base salary ($20.00 × 8 h) paid by their home company. Overtime pay for dispatched employees is paid by the client company. Candidate workers who are on paid leave to commute to school are paid only their base salary by their home company.

3 Simulation Configurations This simulation is based on Japanese society. Therefore, we used a Japanese calendar. The number of employees in the companies was set at a uniformly distributed random value from 51 to 100. In addition, the number of participating companies was 100. The number of issued tokens was obtained by multiplying the number of employees taking leave by 12.

46

K. Abe et al.

3.1 Motivation Value The threshold for the motivation value was set to 2σ so that the top 2.3% of the population of employees attend school. The reasons for this percentage are the rate of working adults in Japan who receive education at graduate schools or other institutions is 2.4% [27], and 2.4% ≈ 1.977σ, 2.3% = 1.995σ so that we can make threshold value simpler. The initial motivation value is the same for both candidate workers and normal workers. However, candidate workers are concerned that they may lose their jobs because of AI or machines in the near future. Therefore, in the proposed token exchange system, their motivation value is increased by 20, because candidate workers expect that they will be able to go to school more easily. In this experiment, the motivation value is average = 50 and 1σ = 10. Thanks to implement of the token/employee exchange system, we expected the candidate workers’ situation to approach the 2.4% (≈ 2σ) that includes working students. Therefore, we set the threshold 70. The timing of the motivation value is increased is the day before the first day of the fiscal year.

3.2 Simulation Period The simulation period was considered from April 1, 2022, to March 31, 2027. There were two reasons for this. First, the Japanese fiscal year extends from April 1 to March 31. Second, we considered a 10-year period in a preliminary experiment, and the number of commuting employees dropped to almost zero in the fifth to sixth years. This was the case because a certain number of employees had a low motivation value initially, so even if the number increases by 20 after the introduction of the system, the threshold value of 70 is not reached.

3.3 Amount of Work and Peak/Off-Peak Season The amount of work per day (TW ) for each company (i) is calculated by (1). The annual calendar is divided into quarters: one peak season, one off-peak season, and two normal seasons. These periods vary from company to company and are randomly set by uniform distribution. T Wi = I Wi × random(1.1, 0.9) + C Wi + P Wi

(1)

IW = initial amount of workload ([total population of each company] × 8); CW = carryover workload from yesterday; PW = IW i × 1.1 (peak season) or IW i × 0.9 (off-peak season).

How Can Adults Make Time to Study: A System for Employee Sharing …

47

3.4 Token/Employee Exchange and Paid Leave Rate The following three patterns were conducted to compare the effects of the token/ employee exchange. Note that regardless of whether employees take extended leave for study, those whose motivation value is above 70 from the beginning and who have more than 2.5 h of free time attend school after work. In this study, the paid leave employees are clerical or other less specialized positions based on prior research showing that education is negatively correlated with the likelihood of computerization [1] and the prediction that probability of computerization is extremely high for general administrative clerks [2]. Therefore, employees who are lent to other companies do not need special skills, knowledge, or experience. In addition, the setting on the simulator in this study does not configure willingness of the employees who are lent. • Simulation 1: Implementing a Token/Employee Exchange with Several Employees Attending School Each Year. The company grants leave to a few of the candidate workers. The maximum number is 2% of the total number of employees. This number is rounded to the nearest whole number. • Simulation 2: Several Employees Attending School Without a Token/ Employee Exchange. The conditions under which several employees attend school each year are the same as in (1). In other words, the simulation assumes that each company has several vacancies each year that are not filled. • Simulation 3: No Token/Employee Exchange and No Vacancies. In this simulation, companies are assumed to have no excess or shortage of employees. No workers take extended leave to go to school, and the token/employee exchange is not implemented.

4 Matching In this system, employees are borrowed every month. Matching for this purpose is performed on the last Saturday of each month based on the current month’s data. Then, the employees and tokens are exchanged the following month.

4.1 Scoring of Workloads and Calculation of Required Number of Employees To match companies with each other, each company scores its workload. Each company calculates whether its daily carryover workload is greater than 2.5% of its normal workload for the business day from the first to the last Saturday. If it is greater than 2.5%, it records a score of + 1; if it is between 0 and 2.5%, it records a score of 0; if it is less than 0%, it records a score of − 1. Note that 2.5% is the average

48

K. Abe et al.

amount of work carried over if there were no excess or shortage of employees, as calculated in the preliminary experiment. The workload scores are totaled and reported to the government on the last Saturday of each month. At this time, each company’s score is subtracted from the number calculated by multiplying the number of employees by 0.0001 to differentiate rankings by employee size in the event of identical scores. As a result, the company with the highest score as of the last Saturday of each month is considered the “busiest company” for that month. Companies with a score of 1 or higher report to the government the number of employees they would like to borrow for the following month. That number is calculated based on how many more employees would reduce the amount of work carried over to 2.5% or less. If the company knows that the following month is a peak season, it reports this number plus an additional number equal to the number of employees taking extended leave.

4.2 Calculation of the Number of Employees Available to Borrow Each company also calculates how many employees can be lent to other companies if each person works overtime for no more than 30 min. The number of employees calculated here is reported to the government on the last Saturday of each month as the number of employees that can be lent to other companies in the following month. However, if the next month is a peak season, no employees are lent. If the next month is an off-peak season, the number of employees available to lend increases by an additional 10%.

4.3 Matching Between Borrowing and Lending Companies Matching is performed on a one-to-one basis between “the company that was the busiest that month” and “the company that can lend the most employees.” Then, in turn, the second company matches with the second company, the third company matches with the third company, and so on. If there is a discrepancy between the number of companies that want to borrow and those that lend the lender takes priority. A borrowing company must have sufficient tokens to borrow workers. If the number of people available for lending by a lending company is 0, the company does not receive any tokens or lend any people.

How Can Adults Make Time to Study: A System for Employee Sharing …

49

5 Definition of Completion of Education Employees attending school are considered to have completed their education when they have completed a minimum of 1200 h of in-school study. This must be accomplished within 1 year of being on leave for candidate workers who take extended paid leave through this system. For employees who do not take paid leave but attend school after work, this may be accomplished across multiple fiscal years. In both cases, the school does not take place over summer and winter breaks, which are assumed to occur from July 1 to September 30 and from February 1—March 31.

6 Results and Discussion The results of the simulations are shown in Fig. 2 and Tables 2, 3, 4 and 5. Note that these are averages of the results of 10 simulations. The random seed was fixed for each simulation run.

Fig. 2 Examples of daily workload (simulation 1: a, simulation 2: b, simulation 3: c)

Table 2 Entire population Item

Average population

Average population per company

Population of workers

7548/100 companies

74.58/1 company

Candidate workers

1129/100 companies

11.29/1 company

Table 3 Education complements for simulation 1 Item

Average population

Average population per company

Workers who completed a course of study

718.4/100 companies 7.18/1 company

Candidate workers who completed a 568.4/100 companies 5.68/1 company course of study

50

K. Abe et al.

Table 4 Education complements for simulation 3 Item

Average population

Workers who completed a course of study

172.5/100 companies 1.73/1 company

Candidate workers who completed a 0/100 companies course of study

Average population per company

0/1 company

Table 5 Sum of the cost for time and payment in 5 years Simulation

Overtime (h)

Payment ($)

Remark

Simulation 1

3273304.55

1566850496.62

Includes payroll for school leave

Simulation 2

3476635.86

1570498008.44

Includes payroll for school leave

Simulation 3

2738412.54

1553583660.93

$41165.44/1 worker in 1 year

6.1 Effectiveness of Completing Education Tables 2 and 3 show that of an average of 11.29 candidate workers per company, 5.68 could complete their schooling over 5 years. In contrast, Table 4 shows that none of the candidate workers could complete their education. This result indicates that it was difficult for them to find time to study after work on their own. This result confirms the effectiveness of the system proposed in this study in facilitating the completion of schooling by candidate workers.

6.2 Effectiveness of Business Execution Figure 2a–c) each shows the daily workload over the 5 years. One of the ten simulations is shown as an example; the other nine showed similar trends. In Fig. 2c, the workload remains flat because there is no excess or shortage of employees. Similarly, Fig. 2a shows a relatively flat trend. This result indicates that no extreme workload carryover occurred. In contrast, Fig. 2b shows the result when there were vacancies that were not filled by employee exchanges. It may be observed that the workload accumulates until the number of candidate workers whose motivation value exceeds the threshold is reduced. The results show that our proposed token/employee exchange system allows workers to perform their daily tasks properly.

How Can Adults Make Time to Study: A System for Employee Sharing …

51

6.3 Which is Less Expensive, Token/Employee Exchange or New Hire? Generally, employers would hire a new employee or contract a temporary worker through a staffing agency rather than exchange employees between companies. If staffing through employee exchanges is sufficiently inexpensive, it may be more economical than hiring new people. Here, the average increase in payroll spending per company over 5 years with token/employee exchanges, net of government subsidies, is as follows. 13266835.69/100 = $132668.36 Furthermore, from Table 3, because there are 5.68 vacancies in 5 years, the amount each firm would have spent to fill each vacancy is as follows: 132668.36/5.68 = $23357.11 This is cheaper than the annual salary of $41165.44/year (Table 5) if there is no excess or shortage of employees. Therefore, an employee exchange is less expensive than hiring a new employee to fill a vacancy.

6.4 Compensation to Companies that Had Many Opportunities to Lend Employees In our proposed system, token circulation is also crucial. Participating companies can either borrow or lend employees in a given year. This means the company can use the tokens it earns when it lends an employee and can exercise those tokens when short-staffed. The adequacy of token distribution can be determined by examining the remaining number of tokens at each company at the end of the fiscal year; if the difference between them is small, it means that the number of tokens spent and received was approximately the same. Because this is synonymous with the number of employees borrowed and lent, it is also an indicator of whether the burden on each company is fair. The number of companies for which the difference between the number of tokens at the beginning of the year and the actual number of tokens at the end of the year was greater than 1 was 32.00 (standard deviation 1.00) on average for the 10 simulations, and the total number of difference in tokens was 181.70 (standard deviation 14.59). If the government compensated a company whose end-of-year token count was greater than 110% with an amount equal to half its hourly wage ($10.00, 8 h/day, 20 business days/month), the government’s per-company expenditure would be as follows. (181.70 × 10.00 × 8 × 20)/32.00 = $9085.00

52

K. Abe et al.

Table 6 The average number of employees lent at the end of the year in simulation 1 Item Average (people) Standard deviation Subsidies ($)

Year 1 63.40

Year 2 58.00

Year 3 58.00

Year 4 20.40

Year 5 4.80

12.46

13.51

14.75

11.51

6.20

202880.00

185600.00

185600.00

65280.00

15360.00

Thus, the subsidy to be disbursed by the government was $654,720 over 5 years.

6.5 Government Subsidies for Companies that Loaned Employees at the End of the Fiscal Year Under this system, the government compensates companies that lend out employees in the last month of the fiscal year; tokens also expire on the last day of the fiscal year. In other words, a company that lends out employees in the last month of the fiscal year cannot use the tokens they received at that time. This should be compensated by the government in cash. Table 6 shows the number of employees lent in March for 10 simulations. Note that the subsidy was calculated as $20/hour, 8 h/day, and 20 days per month.

7 Conclusion In this study, we propose a method to enable working people to secure study time at school and a method for filling employee vacancies at a lower cost than hiring new people through a token/employee exchange system. The results of a simulation have confirmed that both of these goals can be effectively realized with this approach. Even if AI and machines do not directly replace human jobs, new technologies are continuously being developed. Therefore, we believe that this system should be considered as one method for workers to secure time to study while retaining their employment. Further work should be conducted to present results with greater precision and simulation results under different conditions. For example, feasibility and costs should be calculated for a limited number of countries, regions, and job categories. In addition, if staffing agencies participate, changes in costs must be investigated. Acknowledgements This work was supported by JSPS KAKENHI Grant No. 21K12144, by JST AIP Trilateral AI Research Grant No. JPMJCR20G4, and by JST Mirai Program Grant No. JPMJMI20B3.

How Can Adults Make Time to Study: A System for Employee Sharing …

53

References 1. Frey, C.B., Osbone, M.: The future of employment: how susceptible are jobs to computerisation? Technol. Forecast. Soc. Chang. 114, 254–280 (2017) 2. Frey, C.B., Osbone, M.: Computerization and the future of work in Japan. Nomura Research Institute, Ltd., https://www.nri.com/-/media/Corporate/jp/Files/PDF/journal/2017/05/01J.pdf. Accessed 14 June 2022 3. Nomura Research Institute, Ltd.: 49% of Japan’s workforce can be replaced by artificial intelligence, robots, etc. https://www.nri.com/-/media/Corporate/jp/Files/PDF/news/newsrelease/cc/ 2015/151202_1.pdf. Accessed 07 June 2022 4. McKinsey & Company.: The future of work in Japan, https://www.mckinsey.com/jp/~/media/ McKinsey/Locations/Asia/Japan/Our%20Insights/Future%20of%20work%20in%20Japan/ Future%20of%20work%20in%20Japan_v3_jp.pdf. Accessed 14 June 2022 5. Petropoulos, G., Pichler, D., Chiacchio, F.: The Impact of Industrial Robots on EU Employment and Wages: A Local Labour Market Approach, Bruegel Working Paper, vol. 2018/02 (2018) 6. Yarlagadda, R.T.: Future of Robots AI and automation in the United States. Int. Multidisc. J. 1(5), 6 (2015) 7. Fleming, P.: Robots and organization studies: why robots might not want to steal your job. Organ. Stud. 40(1) (2018) 8. Huang, M., Rust, R.T.: Artificial intelligence in service. J. Serv. Res. 21(2), 155–172 (2018) 9. Hmoud, B., Laszlo, V.: Will artificial intelligence take over human resources recruitment and selection? Netw. Intell. Stud. 13, 21–30 (2019) 10. Ministry of Health, Labour and Welfare of Japan Homepage. https://www.mhlw.go.jp/stf/sei sakunitsuite/bunya/koyou_roudou/koyou/kyufukin/d01-1.html. Accessed 25 May 2022 11. Statistics Bureau of Japan Homepage. https://www.stat.go.jp/data/roudou/sokuhou/nen/ft/ index.html. Accessed 24 May 2022 12. Ministry of Health, Labour and Welfare of Japan Homepage. https://www.mhlw.go.jp/stf/hou dou/newpage_19368.html. Accessed 24 May 2022 13. Cosculluela, J.C.: A set of patterns in adult education. In: 23rd European Conference on Pattern Languages of Programs, pp. 1–12. Association for Computing Machinery, New York, NY, United States (2018) 14. Kang, S., Cha, J., Ban, S.: A study on utilization strategy of edu-tech-based MOOC for lifelong learning in the fourth industrial revolution. In: 2018 Conference on Research in Adaptive and Convergent Systems, pp. 324–325. Association for Computing Machinery, New York, NY, United States (2018) 15. Thang, L.L., Lim, E., Tan, S.L.: Lifelong learning and productive aging among the babyboomers in Singapore. Soc. Sci. Med. 229, 41–49 (2019) 16. Sarsa, S., Hellas, A., Leinonen, J.: Who continues in a series of lifelong learning courses? In: 27th ACM Conference on Innovation and Technology in Computer Science Education, vol. 1, pp. 47–53. Association for Computing Machinery, New York, NY, United States (2022) 17. Tang, W., Liu, X.: Trends of human resource management in the shared economy era. In: 1st International Conference on Business, Economics, Management Science, pp. 568–570. Atlantis Press, Hangzhou, China (2019) 18. Chen, Z.: Sharing employee: B2B employment model in the era of coronavirus disease 2019 and implication for human resource management. Front. Psychol. 12, 714704 (2021) 19. Lili, Z.: An inverse optimization model for human resource allocation problem considering competency disadvantage structure. Proc. Comput. Sci. 112, 1611–1622 (2017) 20. Song, X., Liu, J., Cao, Y., Long, H.: Optimization of human resource allocation problem considering costs reduction and balance in Chinese state-owned enterprise. In: 4th International Conference on Management Engineering, Software Engineering and Service Sciences, pp. 211– 215. Association for Computing Machinery, New York, NY, United States (2020) 21. Filho, C.F.F.C., Rocha, D.A.R., Costa, M.G.F., Pereira, W.C.A.: Using constraint satisfaction problem approach to solve human resource allocation problems in cooperative health services. Exp. Syst. Appl. 39(1), 385–394 (2012)

54

K. Abe et al.

22. Chien, C., Chen, W., Hsu, S.: An indirect workforce (re)allocation model for semiconductor manufacturing. In: 2008 Winter Simulation Conference, pp. 2201–2208, Winter Simulation Conference, Miami, Florida, United States (2008) 23. Saito, M., Hayashi, H.: P2P human-resource sharing and its redistribution strategy of stable coin. Int. J. Smart Comput. Artif. Intell. 4(2), 1–18 (2020) 24. Hayashi, H.: Comparing decentralized algorithms for dynamic task sharing. In: International Conference on Web Intelligence and Intelligent Agent Technology, pp. 398–405. Association for Computing Machinery, New York, NY, United States (2022) 25. Hayashi, H.: Cost-aware dynamic task sharing among decentralized autonomous agents: towards dynamic patient sharing among hospitals. In: Agents and Multi-Agent Systems: Technologies and Applications 2022, pp. 13–25. Springer Singapore, Gateway East, Singapore (2022) 26. She, Y., Leung, H.: An adaptive strategy for allocation of resources with gradually or abruptly changing capacities. In: 2008 20th IEEE International Conference on Tools with Artificial Intelligence, pp. 415–422. IEEE, Dayton, Ohio, United States (2008) 27. Gender Equality Bureau, Cabinet Office of Japan Homepage. https://www.gender.go.jp/about_ danjo/whitepaper/r01/zentai/html/honpen/b1_s00_03.html. Accessed 25 Aug 2022

Data Discretization for Data Stream Mining Anis Cherfi and Kaouther Nouira

Abstract Data discretization has become essential task to cope with streaming data. However, this field still requires particular attention. Indeed, various constraints should be taken into consideration to cope with discretization in data stream scenarios. In this paper, we discuss the discretization process for data stream scenarios. In this setting, we examine the proposed algorithms in the literature, their limitations and the new challenges. Also, we propose a novel discretization method so called Hybrid Online Discretization (HOD). Based on the well-known Minimum Description Length Principle (MDLP) discretization combined with the Online ChiMerge Algorithm, the proposed method guarantees an efficient and effective discretization. Empirical trials show that in most cases the proposed algorithm outperforms its competitors in terms of accuracy level and reduction rate. Finally, we experimentally demonstrate that coupled with HOD, Naive Bayes classifier shows high classification accuracy.

1 Introduction: Motivations and Challenges Many studies have proven that the success of Data Mining task depends heavily on information bias and data quality. In many emerging real-word problems we can assume that we will deal with low-quality mixed data. Raw quantitative data usually comes with many drawbacks which demands cleaning, transformation, and discretization before the learning process [1–4]. Low data quality and imperfections such as redundancies, inconsistencies, and noise have a direct impact on data mining task results. Therefore, improving data quality is crucial to obtain useful patterns. There are many evidences prove that data discretization techniques minimize the A. Cherfi (B) ESPRIT School of Business, Tunis, Tunisia e-mail: [email protected] Data Science and AI, Esprit School of Engineering, Tunis, Tunisia K. Nouira Université de Tunis, ISGT, LR99ES04 BESTMOD, 2000 Le Bardo, Tunisia © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_5

55

56

A. Cherfi and K. Nouira

information bias and maximize the data quality which improve the performances and effectiveness of data mining algorithms [5–7]. Moreover, discretization is used to reduce irrelevant, noisy and redundant instances. Thus, redundant values are combined in an interval. Discretization corrects inconsistencies in the data. By minimizing the number of intervals, discretization reduces the amount of data with high noise content which contains conflicting instances. The benefits of feature discretization include a reduction in inconsistency level, discarding the noisy, irrelevant and redundant information present in data, while improving the generalization power of data [8–10]. Reducing the cardinality of the attribute improve both data quality and complexity. Even for most of data mining algorithms that work directly with numerical attributes, discretization as a preprocessing task can improve the model performances. The achievement of algorithms belonging to the set of statistical learning and those that use information measures ( i.e., Support Vector Machine (SVM), K-nearest Neighbor (KNN)...) is dramatically affected by attribute discretization. Good quality solutions and higher performances are observed in the results obtained on discretized data [11–14]. Many researchers prove that discretization improves the performances of data mining algorithms since it gives a learning algorithm a smaller space to reason about, with more examples in each part of the space [15, 16]. Also discretization performs features selection by mapping the feature values into a single interval [3, 17]. Nowadays, it is surprising that discretization methods for data streams remain fairly unexplored and have received little attention [7, 18]. The main reason may be that many constraints should be taken into consideration while discretizing data to guarantee positive results. First of all, handling the appearance of concept drifts is one of the major keys for discretization success. That means that at some point the relation between the input and the target variable or the input data statistical characteristics may change [19]. The nature of data streams is based on the continuous change of attribute values over time. Therefore, in discretization process, the number of intervals and the definition of boundary points may change as the stream progress. Thus, results follow shifts in data characteristics which results to discretized values whose meaning changes over time. While some authors argue that this can bias users against using discretization. Supported by [18], we agree that change in data values is one of the basics meanings of data streams. Therefore, to maintain the relevant meaning of input data, discretization methods might smoothly adapt intervals and so change values meaning. Data streams discretization in the literature requires various level of refinement where sophisticated iterations are performed using several data structure to smoothly adapt the discretization to concept drift. Here we outline that one of the main challenges is to improve intervals and adapt discretization scheme quality without involving increased computational cost [7]. Moreover, performances of classification and prediction models are directly influenced by the number of intervals produced by discretization methods. Discretization methods reduce quantitative data space considering a tradeoff between minimizing the number of intervals and maximizing data consistency. Also, by reducing the number of intervals, discretization methods can eliminate some irrelevant attributes. It performs feature selection by discretizing a continuous attribute into one interval.

Data Discretization for Data Stream Mining

57

This major advantage in traditional discretization methods presents a new challenge in data stream scenarios. Actually, the proposed discretization methods for data streams require a predefined number of intervals which omit the possibility of eliminating irrelevant attributes. In the next section we discuss with more details the advantages and challenges of discretizers for data streams from a theoretical and practical point of view. This paper is organized as follows: The related and advanced works on discretization for data streams are provided in Sect. 2. In Sect. 3, we extend more details about the new discretization algorithms we implement. Section 4 presents experiment evaluation of the algorithm in real-world problem and we discuss the results. Finally Last section concludes the paper and presents some future research lines.

2 Related Works There are rare examples of discretizers were designed specifically to operate on data streams environments for machine learning purposes. This is explained by the fact that many requirements must be taken into consideration while designing a discretization method for data stream scenarios. Table 1 lists the most relevant discretization methods for data streams. Partition Incremental Discretization algorithm (PiD) is one of the first attempts to deal with data discretization for streaming environments [20]. According to [21, 22], equal-frequency can be used as an alternative strategy to improve this discretizer. Whereas Incremental Flexible Frequency Discretization (IFFD) requires a random ordered streaming records arriving, which is impossible in many learning problems [7]. When it comes to Optimal Flexible Frequency Discretization (OFFD) [23] , authors report that is not optimal in all the cases [24]. In the case of Online ChiMerge based algorithm (OCM) [25], it does not require a user-provided number of intervals to initialize the discretization process, yet using several data structures may prevent OCM algorithm to be used in limited computational resources problems [7, 25]. According to a recent study, Incremental Discretization Algorithm (IDA) is the agilest and the most effective discretizer [7]. To conclude, in this paper we propose a supervised discretization algorithm based on OCM discretizer. As discussed previously, OCM is the only supervised algorithm proposed in the literature. Based on critics summarized in this section, our objective is to avoid user defined parameters related to number of intervals, and to propose a data driven discretization algorithm. We aim to improve the trade-off between accuracy level and reduction rate. The proposed algorithm is discussed with more details in the next section.

58

A. Cherfi and K. Nouira

Table 1 Summary description of data streams discretization algorithms Discretizer Layers PiD unsupervised (2006) IFFD unsupervised (2006)

Splitting Merging Splitting Splitting

OCM supervised (2012) IDA unsupervised (2014)

Splitting Merging Splitting Update

Equal-width Recursive entropy Equal-Frequency Fix Frequency Discretization Set new values as cut points ChiMerge Discretization It approximates quantiles It updates intervals statistics

# intervals=200 # intervals=40 # min intervals=30 # max intervals=60 initial Elements=100 Sample size = 1000 num intervals=5

Fig. 1 HOD discretization layers

3 Hybrid Online Discretization In this section, we introduce the Hybrid Online Discretization (HOD). The HOD algorithm is composed by two layers to obtain an optimal discretization scheme. The first layer splits data by creating new cut-points based on MDLP algorithm [26]. In the second layer OCM algorithm is applied to simplify the discretization scheme produced by the previous layer. Figure 1 explains the discretization process In HOD algorithm. The first layer initializes the discretization scheme by reading x examples from DataSource. The input for this phase is x the number of instances to create the first discretization scheme. The x examples are used to identify the number of intervals in each attribute. The algorithm retains only distinct attribute values to form a Candidate Cut Points (CCP) list. A CCP must be a value between two examples having different class values, otherwise it is rejected. As described in Algorithm 1, we apply the MDLP discretization to obtain the initial discretization scheme. We introduce feature selection advantages in discretization for data streams by using MDLPC criterion. The resulting discretization scheme can contain only one interval. Thus, the attribute can be omitted. Each time we observe a new value of the random examples, layer 1 inserts the new value in the list of intervals to decide if a new CCP will be created or not.

Data Discretization for Data Stream Mining

59

Fig. 2 Intervals update process

Definition 1 A value of the random examples is defined by the attribute value Vi and a class value Ci , belong to I j , where I j is an interval defined by the majority class C j and two Cut Points (CP) C P j and C P j+1 . Vi is selected as a CCP for interval I j if C P j ≤ Vi < C P j+1 and Ci = C j . Theorem 1 If Vi belongs to the minority class of an interval, then Vi is a CCP. Proof As described in Fig. 2, assume we have an interval I2 with a majority class C2 and two born B1 and B2 . The information entropy Ent(I2 ) is minimal. If a new value Vi with a class value Ci belong to the interval I2 (B1 ≤ Vi < B2 ) and Ci = C2 then the information entropy Ent(I2 ) will remain minimal if we insert the new value in I2 . Else, if Ci = C j then adding the new value in interval I2 will increase the information entropy Ent (I2 ). Thus a binary discretization for I2 defined by the new value Vi may reduce the interval impurity and information entropy. If the new arriving value is classified in the right interval and has the same label as this interval, then, no changes will be performed. Else the arriving value will be selected as a CCP. The second layer starts by initializing a balanced binary search tree using results from MDLP discretizer and the discretization scheme is updated only periodically. The algorithm constructs the initial intervals set which depends on the data distribution. Then, the algorithm performs merging process based on initialized statistics and the set of intervals from layer 1. Interval merging is performed Algorithm 1 Layer 1(n, CC P, I nter vals) 1: Inputs: n: the number of instances to initialize the model; CC P: the candidate cut point list; I nter vals: the curent list of intervals; 2: Outputs: Layer : the index of the current Layer; CC P: the updated candidate cut point list; 3: While E ← DataSour ce() = null do 4: if (count < n) then Data ← E; 5: if (count == n) then 6: CC P ← M DL P(Data); 7: Layer ← 2; 8: end 9: if (count > n) then 10: r es ← I nser t (E, I nter vals); 11: if Not (r es) then 12: Layer ← 2; 13: CC P ← E; 14: end 15: end 16: count ← count + 1; 17: end

60

A. Cherfi and K. Nouira

Table 2 Summary description of data streams Dataset Src. #Inst. #At. Airlines CovtypeNorm ElecNormNew HyperPlane Kddcup_10 Poker-lsn Power_supply Spambase

[27] [27] [27] [27] [28] [27] [29] [30]

500,000 581.011 45.311 100,000 494.020 829.201 29.928 4601

20 54 8 10 41 10 2 57

#Nu.

#No.

#Cl.

20 10 7 10 39 5 2 57

0 44 1 0 2 5 0 0

4 7 2 5 2 10 24 2

as in batch ChiMerge without stopping the processing to update directly the tree with new arriving examples. Finally, the algorithm updates the current discretization scheme with temporary intervals generated in the current layer. Each time a new CCP is selected, in layer 2, the algorithm applies the OCM method to investigate the usefulness of creating a new interval based on the new value. The binary search tree used by OCM discretizer is updated using only CCP to reduce algorithm complexity.

4 Experimental Set up In this section, experimental comparisons are conducted on 8 datasets for mining data streams. In order to investigate the discretization performances, we evaluate the algorithms from different perspectives. We carried out experiments on 8 datasets taken from different sources, summarized in Table 2. HOD discretizer has been integrated in MOAReduction library as a part of MOA software. All experiments have been executed in a single machine with a processor Intel Xeon E3-1230 (Cores: 4 × 3.30 GHz (Single Quad Core)) and 16 GB of RAM. Also, many parameters should be tuned to grantee the success of discretization process. Parameters, and other considerations are summarized in the following: • Discretization parameters: initial elements = 100, window size = 1 (default). • PiD parameters: α = 0.755, initial bins = 500, instances to update layer #2 = 10,000, min = 0, and max = 1. • NB parameters: initial elements = 100, window size = 1 (default)

Data Discretization for Data Stream Mining

61

5 Analysis and Empirical Results To evaluate the ability of the proposed algorithm to reduce the continuous attribute space without losing important information, we use 3 discretization methods (IDA, OC, and PiD) as recommended in [7]. All of the algorithms are used to discretize data. Then, to investigate the quality of discretization schemes, NB classifier is used to train and test data with and without explicit discretization. Accuracy and number of intervals generated by discretization algorithms are summarized in Table 3. From these results, we can conclude that: • The best discretization algorithm in average is the IDA discretizer. It outperforms the set of used algorithms in both accuracy and number of intervals. It shows a good combination of accuracy level and reduction rate especially in the cases of relatively small datasets (electNormNew, hyperPlane, power_supply, and spambase). However, IDA results are pretty close to those obtained by the proposed discretizer HOD. • The proposed algorithm HOD is the closest competitor of IDA algorithm with 2.4 units bellow in term of accuracy level. However, it obtains the best accuracy mark in 4/8 datasets. In three cases, HOD algorithm yields less intervals than IDA, in two of these cases HOD has the best accuracy. In average it is the secondbest algorithm for both accuracy level and reduction rate. Also, it is specially remarkable that HOD obtains an outstanding mark in the poker-lsn dataset. • Regarding reduction rate from Fig. 3c and Table 3, OC and PiD yield the worst results. For OC, both accuracy and reduction rate are less competitive. In average, the OC accuracy results are very close to those obtained by the base solution. PiD discretizer generate the most complicated discretization schemes, but it yields more competitive accuracy results than OC and NB. • As expected, in all cases, NB test accuracy with discretized data outperforms the base solution (NB classification without explicit discretization). Figure 3a,b depicts CPU usage and memory usage. As discussed before, OC discretizer combines a high number of data structure and sophisticated process which

Fig. 3 a, b CPU processing time (in s.) and memory usage (in RAM-hours) over the data stream progress. c Discretization reduction

62

A. Cherfi and K. Nouira

Table 3 Classification test accuracy and number of intervals after discretization Data set HOD IDA OC PiD

Airlines CovtypeNorm ElectNormNew Hyper Kddcup_10 Poker-lsn Power_supply Spambase MEAN

Acc

#Inter

Acc

#Inter

Acc

#Inter

Acc

#Inter

Naive Bayes Acc

65.29 76.39 76.38 38.96 99.78 69.68 04.48 93.80 65.59

67 80 153 70 216 22 12 469 136

64.40 76.12 76.06 47.71 99.55 68.95 13.62 97.67 68.01

48 330 54 66 154 55 18 198 115

65.29 73.61 74.12 36.78 99.77 56.70 05.15 96.41 63.48

62 99 164 82 300 19 13 658 174

64.24 73.25 76.87 51.21 99.55 56.49 04.00 97.18 65.35

20 354 97 110 464 56 418 320 229

64.54 60.42 72.96 47.26 99.66 59.44 16.28 82.02 62.82

explain the results obtained in Fig. 3a,b. IDA discretizer always performs better than others. Its results are pretty close to those obtained by NB. While detailed results show that memory usage of HOD algorithm is too far than OC and PiD. Also, for CPU usage, HOD is more competitive, it performs faster than OC and PiD, and slower than IDA and the base solution, but it still acceptable.

6 Conclusion In this paper we have summarized the key issues that surround data discretization for data stream scenarios. Most proposed algorithms require various level of refinement and sophisticated iterations to perform discretization. The high number of parameters that should be tuned to guarantee better discretization scheme represent the major disadvantage of proposed discretization algorithms. The new algorithm proposed in this paper, aims to detect the number of intervals based on MDLP criterion. HOD compute the initial number of intervals and performs discretization based on MDLP discretizer and OCM algorithm. Results show that NB classifier combined with HOD discretizer generates efficient and effective models. In many cases, HOD outperforms his competitors in both accuracy level and reduction rate. We conducted our experiments using NB as a base solution. And we have shown that HOD algorithm improves the results of this classifier. Future research could explore the benefits of the proposed discretization algorithm in contexts of decision trees and clustering for data stream scenarios.

Data Discretization for Data Stream Mining

63

References 1. Garcia, S., Luengo, J., Sáez, J.A., Lopez, V., Herrera, F.: A survey of discretization techniques: taxonomy and empirical analysis in supervised learning. IEEE Trans. Knowl. Data Eng. 25(4), 734–750 (2013) 2. García, S., Luengo, J., Herrera, F.: Data Preprocessing in Data Mining, pp. 59–139. Springer, New York (2015) 3. Wang, B., Zhang, J., Zhang, Z., Luo, W., Xia, D.: Traffic identification in big internet data. In: Big Data Concepts, Theories, and Applications, pp. 129–156. Springer International Publishing (2016) 4. García-Borroto, M., Martínez-Trinidad, J.F., Carrasco-Ochoa, J.A.: A survey of emerging patterns for supervised classification. Artific. Intell. Rev. 42(4), 705 (2014) 5. Tahan, M. H., Asadi, S.: MEMOD: a novel multivariate evolutionary multi-objective discretization. Soft Comput., 1–23 (2017) 6. García, S., Ramírez-Gallego, S., Luengo, J., Benítez, J.M., Herrera, F.: Big data preprocessing: methods and prospects. Big Data Anal. 1(1), 9 (2016) 7. Ramírez-Gallego, S., Krawczyk, B., García, S., Wo´zniak, M., Herrera, F.: A survey on data preprocessing for data stream mining: current status and future directions. Neurocomputing 239, 39–57 (2017) 8. Cherfi, A., Nouira, K., Ferchichi, A.: Very fast C4.5 decision tree algorithm. Appl. Artific. Intell. 32(2), 119–137 (2018) 9. Cano, A., Nguyen, D.T., Ventura, S., Cios, K.J.: ur-CAIM: improved CAIM discretization for unbalanced and balanced data. Soft Comput. 20(1), 173–188 (2016) 10. Ramírez-Gallego, S., García, S., Mouriño-Talín, H., Martínez-Rego, D., Bolón-Canedo, V., Alonso-Betanzos, A., Herrera, F. et al.: Distributed entropy minimization discretizer for big data analysis under apache spark. In: Trustcom/BigDataSE/ISPA, vol. 2. 2015 IEEE, pp. 33–40. IEEE, (2015) 11. Jiang, F., Sui, Y.: A novel approach for discretization of continuous attributes in rough set theory. Knowl.-Based Syst. 73, 324–334 (2015) 12. Ferreira, A.J., Figueiredo, M.A.: Incremental filter and wrapper approaches for feature discretization. Neurocomputing 123, 60–74 (2014) 13. Ramírez-Gallego, S., García, S., Mouriño-Talín, H., Martínez-Rego, D., Bolón-Canedo, V., Alonso-Betanzos, A., Herrera, F.: Data discretization: taxonomy and big data challenge. Wiley Interdis. Rev.: Data Min. Knowl. Discovery, 6(1), 5–21 (2016) 14. Yang, Y., Webb, G. I., Wu, X.: Discretization Methods. In: Data mining and knowledge discovery handbook, pp. 101–116. Springer, Boston, MA, (2009) 15. Ramírez-Gallego, S., García, S., Benítez, J.M., Herrera, F.: A distributed evolutionary multivariate discretizer for Big Data processing on Apache Spark. Swarm Evol. Comput. 38, 240–250 (2018) 16. Zhou, L., Pan, S., Wang, J., Vasilakos, A.V.: Machine learning on big data: opportunities and challenges. Neurocomputing 237, 350–361 (2017) 17. Menzies, T., Mizuno, O., Takagi, Y., Kikuno, T.: Explanation verus performance in data mining: a case study with predicting runaway projects. JSEA 2(4), 221–236 (2009) 18. Webb, G.I.: Contrary to popular belief incremental discretization can be sound, computationally efficient and extremely useful for streaming data. In: Data Mining (ICDM), 2014 IEEE International Conference on, pp. 1031–1036. IEEE (2014) 19. Gama, J., Žliobait˙e, I., Bifet, A., Pechenizkiy, M., Bouchachia, A.: A survey on concept drift adaptation. ACM Comput. Surv. (CSUR) 46(4), 44 (2014) 20. Gama, J., Pinto, C.: Discretization from data streams: applications to histograms and data mining. In: Proceedings of the 2006 ACM symposium on Applied computing, pp. 662–667. ACM (2006) 21. Gama, J.: Knowledge discovery from data streams. CRC Press (2010) 22. Pevný, T.: Loda: lightweight on-line detector of anomalies. Mach. Learn. 102(2), 275–304 (2016)

64

A. Cherfi and K. Nouira

23. Wang, S., Min, F., Wang, Z., Cao, T.: OFFD: Optimal flexible frequency discretization for naive bayes classification. Adv. Data Min. Appl., 704–712 (2009) 24. Rajappan, S., Rangasamy, D.: Estimation of incomplete values in heterogeneous attribute large datasets using discretized Bayesian max-min ant colony optimization. Knowl. Inf. Syst., 1–26 (2017) 25. Lehtinen, P., Saarela, M., Elomaa, T.: Online Chimerge Algorithm, pp. 199–216. Foundations and Intelligent Paradigms, Data Mining (2012) 26. Fayyad, U., Irani, K.: Multi-interval discretization of continuous-valued attributes for classification learning (1993) 27. A. Bifet, G. Holmes, R. Kirkby, B. Pfahringer, MOA: massive online analysis. J. Mach. Learn. Res 28. KDUS’s streams repository. http://mlkd.csd.auth.gr/conceptdrift.htm (2020) 29. Zhu, X.: Stream Data Mining Repository. http://www.cse.fau.edu/\LY1\aexqzhu/stream.html (2020) 30. Lichman, M.: UCI machine learning repository. [http://archive.ics.uci.edu/ml]. Irvine, CA, University of California, School of Information and Computer Science (2013)

Linear Machine Learning Algorithm for Early Annual Corn Yield Prediction Ivan Kralj, Mario Kusek, and Gordan Jezic

Abstract Machine learning is an important decision support tool for predicting crop yields, including supporting decisions about which crops to grow and what to do during the crop growing season. Various machine learning algorithms have been used to support crop yield prediction research, including K-Nearest Neighbor, Random Forest Classifier, logistic regression and many more. The objective of this paper is to develop a linear machine learning algorithm based on a Generalized Linear Model (GLM) that is more accurate and also provides early prediction of corn yield with a relative error of less than 20%, which is crucial for decisions on allocating harvesting and storage resources in Croatia. Input parameters for our model include various climate and greenhouse gas parameters. We examined how accurate the corn yield prediction would be if farmers wanted to know what the corn yield would be at a given harvest date during the R1 (Silking) phase of the corn and thereafter. For this test case, the relative error for our model was 11.63%, while for GLM it was 12.55%.

1 Introduction Precision agriculture is a new field where new technologies such as artificial intelligence are being used to improve agriculture worldwide. One of the biggest challenges in this field is predicting crop yields. This information is critical for farmers because it can provide information for better crop management, improving profitability, environmental quality, and marketing decisions [1]. Various methods and models can be used to predict annual crop yields, including machine learning (ML) and conventional research methods, such as collecting field I. Kralj (B) · M. Kusek · G. Jezic Internet of Things Laboratory, Faculty of Electrical Engineering and Computing, University of Zagreb, Unska 3, 10000 Zagreb, Croatia e-mail: [email protected] M. Kusek e-mail: [email protected] G. Jezic e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_6

65

66

I. Kralj et al.

data. The difference between ML and conventional research methods is that ML allows for the simultaneous analysis of multiple predictor variables, which is not always possible with conventional research methods. Conventional research methods may focus on one or a few variables at a time, making it more difficult to capture the complex relationships among multiple variables that affect crop yield. In addition, the use of ML algorithms can help improve the accuracy of yield prediction compared to conventional research methods, which may be based on assumptions or past experience. ML algorithms can find patterns in the data that may not be apparent to the human eye or traditional statistical methods. ML is the most commonly used method for predicting annual crop yields. According to [2], where 567 relevant studies were retrieved from 6 electronic databases, of which 50 studies were selected for further analysis, the most commonly used algorithm is Artificial Neural Network. In addition, Convolutional Neural Networks (CNN) is the most commonly used deep learning algorithm in these studies, and the other widely used deep learning algorithms are Long-Short Term Memory (LSTM) and Deep Neural Networks (DNN). However, these deep learning algorithms are used to predict crop yield based on satellite images, not using weather and greenhouse gas as data. For this reason, linear machine learning algorithm based on Generalized Linear Model was chosen for it’s properties and simplicity, since it is computed with a simple weighted sum of variables. This means that it allows us to understand the strength of the relationships between variables. This paper focuses on the development of a linear machine learning algorithm based on Generalized Linear Model, that is going to be more accurate than Generalized Linear Model, moreover, for early prediction of corn yield with a relative error of less than 20%, that outputs weighted parameter values with a minimum relative error using data obtained from https://ourworldindata.org and https://open-meteo. com/en/docs/historical-weather-api. Both weather and greenhouse gas parameters are used to calculate corn yield because they affect the final corn yield [3]. The paper is organized as follows. Section 2 gives an overview of related work. Section 3 presents the system architecture is presented and further information on source data and parameters. Section 4 explains in details how the developed linear machine learning algorithm works and outputs the computed weighted parameter values. Section 5 presents the results of developed linear machine learning algorithm and Generalized Linear Model and the data used to obtain these results. Section 6 concludes the paper and provides future work discussion.

2 Related Work The prediction of the annual crop yield is one of the most important data to predict. It plays an essential role in decision making at global, regional and field levels. Many researchers have already developed various machine learning models and algorithms to predict crop yields based on various parameters such as weather data, chemical data, previous crop yields, etc.

Linear Machine Learning Algorithm for Early Annual …

67

Cedric et al. [4] propose a machine learning-based prediction system to predict the yield of six crops, namely rice, maize, cassava, seed cotton, yam, and banana, at the country level in the area of West African countries throughout the year. They combined climate data, weather data, agricultural yields, and chemical data to help decision makers and farmers predict annual crop yields in their country. They used a decision tree, multivariate logistic regression, and K-nearest neighbor models to build their system. In the end, they concluded that the decision tree model performed well with a coefficient of determination (R2 ) of 95.3% while the k-nearest neighbor model and logistic regression performed with R2 = 93.15% and R2 = 89.78%, respectively. Jayanarayana Reddy and Rudra Kumar [5] provided a survey paper in which they examined various machine learning techniques used in crop yield estimation. They provided a detailed analysis of the accuracy of these machine learning techniques, such as the support vector machine technique, which provides 97.77% accuracy, 96.55%, sensitivity and 99.24% precision for crop yield prediction, and an artificial neural network, which provides a mean squared error of 5.1%, a mean absolute error of 6.4%, and a coefficient of determination of 99%, and a multiple linear regression that provides a mean square error of 9.8%, a mean absolute error of 6.9%, and a coefficient of determination of 89%. Manivasagam et al. [6] proposed a crop yield prediction system based on crop detail using land area and environmental parameters such as rainfall, temperature, humidity, pH, etc. They tested various machine learning algorithms, including decision tree, K-nearest neighbor, and Random Forest Classifier. Their goal was to improve the accuracy of crop yield prediction compared to previous work using Naive Bayes and logistic regression, which had about 87% accuracy. Finally, they were able to increase the accuracy of crop yield prediction from 87 to 95% using random forest algorithms, while the accuracy of the decision tree was 90% and that of the K-Nearest Neighbor was 85%. Cubillas et al. [7] developed an effective early crop yield prediction model that is accessible and easy to use for farmers or farm managers through a web-based application. An olive orchard in the Andalusia region of southern Spain served as the test case. The model was estimated using spatio-temporal training data, such as yield data from eight consecutive years and more than twenty meteorological parameters automatically loaded from public web services and belonging to a weather station near the test farm. Finally, they managed to obtain results for the 2020/21 and 2021/22 seasons, where they showed a relative error of 3.58% and 10.41% respectively. The main contribution of this research is the early prediction of crop yields with relative errors of less than 20%, which is crucial for making decisions on investments in tillage and crop marketing. Overall, crop yield prediction has been massively improved. However, most of these case studies only used existing machine learning algorithms. In this paper, we developed a modified linear machine learning algorithm based on Generalised Linear Model (GLM) to provide early prediction of corn yield with a relative error of less than 20%, which is critical for decisions on crop and storage resource allocation using data from https://ourworldindata.org and https://open-meteo.com/en/ docs/historical-weather-api.

68

I. Kralj et al.

Linear machine learning algorithms, such as GLM, are commonly used to predict early corn yields because they can effectively model the relationship between yield and predictor variables. Including multiple predictor variables improves the accuracy of the prediction. For example, the GLM can be used to model the relationship between yield and variables such as temperature, precipitation, soil moisture, and other relevant environmental factors. In addition, the GLM is a class of models that generalises the linear regression model to account for different types of response variables, including continuous, binary, and count data. In the case of early corn yield prediction, yield is usually a continuous variable, making the GLM an appropriate choice.

3 Obtained Data and Model For the test case, we chose Croatia, Zagreb, as the test location, and for the crop, we decided to use only corn yield data. The data is extracted from 2 different sources: • https://ourworldindata.org—From this website we obtained the annual total corn yield in tons per hectare for each country, the annual total gas CO2 per capita in tons for each country, the annual total methane per capita in tons for each country, and the annual total nitrous oxide per capita in tons. • https://open-meteo.com/en/docs/historical-weather-api—From this website we have obtained hourly weather data for Croatia, Zagreb from January 1st 1992 to December 31st 2020. The reason for choosing these dates is that the corn yield data obtained from ourworldindata.org website contains historical yields from 1992 to 2020. Table 1 lists all parameters taken from this website. In total, we decided to use 15 parameters as input to the linear machine learning algorithm, and we extracted about 311,228 data from these sources. The implemented model is based on purely linear algorithm, because they have a great strength due to their characteristics and simplicity, as they are calculated with a simple weighted sum of the variables. The output data of the model is the annual corn yield in tonnes per hectare. y = α0 x0 + α1 x1 + · · · + αn xn

(1)

Equation 1 is closely related to the GLM, which function mathematically as a weighted sum of features, where the mean of the distribution is assumed using the link function, which can be chosen flexibly depending on the type of outcome. The formula for the GLM can be written as shown in 2, where α0 , α1 , etc. are the weighted parameters, x1 , x2 , etc. are the input variables, and g −1 is the inverse of the link function. The link function g may be different for different types of GLMs, e.g., the logit function (logistic regression) function for logistic regression, the identity function for linear regression, or the log function for Poisson regression. The inverse of the link function is the inverse of the function, that is, the function that cancels

Linear Machine Learning Algorithm for Early Annual …

69

Table 1 Weather parameters list and description Weather parameter Description Temperature 2 m Relative humidity Dew point 2 m

Surface pressure Precipitation Rain Direct radiation Wind speed 10 m Wind direction 10 m Evapotranspiration Soil temperature 0–7 cm Soil moisture 0–7 cm

Air temperature at 2 m above sea level in ◦ C Relative humidity from 2 m above sea level in % Temperature to which the air must be cooled (at constant pressure) to achieve a relative humidity (RH) of 100% from 2 m above sea level in ◦ C Atmospheric pressure exerted by air on the surface of the earth in hPa Total precipitation in mm Total rain in mm Sky radiation in W/m2 Speed of wind at a height of 10 m in km/h Direction of wind at a height of 10 m in ◦ Reference evapotranspiration in mm Temperature of soil at depths between 0 and 7 cm in ◦ C Water content of soil at depths between 0 and 7 cm in “water fraction by volume” (wfv or m3 m−3 )

the link function. For example, if the link function is the identity function (linear regression), the formula can be written as shown in 3, which is the link function used for comparison with our developed linear model and α0 is called the intercept. y = g −1 (α0 + α1 x1 + α2 x2 + · · · + αn xn )

(2)

y = α0 + α1 x1 + α2 x2 + · · · + αn xn

(3)

Since our linear model lacks the inverse of the link function, our linear algorithm can be classified as a simplified version of the GLM algorithm [8]. The system used to predict annual corn yield in tons per hectare consists of weather data sets, chemical data sets, total annual corn yield, and the machine learning algorithm, as shown in Fig. 1.

4 Linear Machine Learning Algorithm Linear machine learning algorithm is a statistical method used for predictive analysis. Based on training data, the learning process calculates a weight for each feature to form a model that can predict or estimate the target value.

70

I. Kralj et al.

Fig. 1 System architecture for corn yield prediction using a linear machine learning algorithm

Using Eq. 1, we aim to find all weighted parameters for all 15 input parameters. To achieve this, we decided to include the maximum error percentage, the maximum and minimum weighted parameter values, and the fixed step value. The maximum error percentage is an allowable margin of error, i.e., if we set the maximum error percentage to 5%, any solution above 105% and below 95% of the actual corn yield will be skipped. The minimum weighted parameter value is chosen to select a starting value for each parameter, while the maximum weighted parameter values are chosen to minimize the program runtime. The fixed step amount is used to calculate the weighted parameters. Before calculating the weighted parameters, the program calculates the corn yield using the maximum weighted parameter. If the difference between the actual and calculated corn yield is greater than 300%, the minimum and maximum weighted parameter values are halved and the process is repeated until a value below 300% is reached. This number was chosen as the most ideal value because a higher percentage would cause the program to run far too long, while a lower percentage would decrease the accuracy of the algorithm. Once the ideal minimum and maximum weighted parameter values are found, the program loops from the minimum weighted parameter value to the maximum weighted parameter value for all parameters. This means that in this case the program will enter 15 loops to find the values for all 15 weighted parameters. To find the weighted parameter value, a fixed step value is added to the weighted parameter. If the calculated corn yield is within in allowable margin of error, the program checks to see if the calculated corn yield is closest to the actual corn yield. If this is the case, the calculated corn yield is saved along with the values of the weighted parameters. If it’s not, the algorithm continues to run until it’s finished. Design flowchart is shown in Fig. 2, while the pseudocode is shown in Fig. 3.

Linear Machine Learning Algorithm for Early Annual …

71

Fig. 2 Flowchart of machine learning algorithm

5 Result We used 75% of the dataset for training, while the remaining 25% were selected for testing. After analysing corn yields for Croatia, the years 1992–2020 were used for training, with the exception of 1995, 1996, 2003, 2006, 2012, and 2019, which were used for testing. The reason for choosing these years for training is the large differences in crop yields. To evaluate the quality of corn yield prediction, we calculated the relative error for each model. For the calculation of weighted parameter values, we used only the data from April 10th to September 30th from these years because

72 Fig. 3 Pseudocode of machine learning algorithm

I. Kralj et al.

Linear Machine Learning Algorithm for Early Annual … Table 2 Weighted parameter values for each parameter Parameter name Value for our model Temperature 2 m Relative humidity Dewpoint 2 m Surface pressure Precipitation Rain Direct radiation Wind speed 10 m Wind direction 10 m Evapotranspiration Soil temperature 0–7 cm Soil moisture 0–7 cm CO2 Methane Nitrous oxide Intercept

2.53 × 10−12 − 4.07 × 10−6 2.301 × 10−10 2.31 × 10−6 1.0023 × 10−4 1.033 × 10−5 − 2.4698 × 10−6 − 1.047 × 10−5 − 2.4698 × 10−6 1.3013 × 10−7 − 9.867 × 10−9 − 4.69867 × 10−8 3.3013 × 10−7 3.3013 × 10−7 − 7.0301 × 10−6 0

73

Value for GLM − 3.175 × 10−4 − 1.7788 × 10−4 9.731 × 10−4 − 5.465 × 10−5 1.1069 × 10−1 − 1.101 × 10−1 1.831 × 10−5 − 7.55 × 10−5 5.799 × 10−6 − 3.5828 × 10−2 − 4.869 × 10−4 − 1.012 × 10−2 6.477 × 10−1 19.1829 − 8.70396 300.804401

we wanted to look only at parameter values from sowing to harvest date. According to [9], the best sowing date for corn is in April, and it takes about 16 weeks for corn to reach maturity. All weather parameters were summed for this period. After running our developed linear machine learning algorithm and the GLM with the identity function as the link function, Table 2 shows the result of the weighted parameter values. Using these weighted parameter values, the test years, and the formulas presented in Sect. 3, the relative error for our model was 11.36%, while it was 13.1% for the GLM. Now that we know the values of each weighted parameter, we can use these values to calculate early corn yield. According to [10], the R1 (Silking) phase of corn is one of the most critical phases in determining yield potential, and the R1 stage begins about 10–12 weeks after corn planting. Based on this information, we decided to evaluate the prediction of early corn yields when the R1 phase begins, which is in August if sowed date was in April. For our test case, we used the 1st day of August. To calculate corn yield, we decided to multiply the weather weighted parameter values. To calculate the multiplier, we can use the formula in in 4, where numOfDaysHarvestDate is the number of days from January 1st to the corn harvest date, numOfDaysSowDate is the number of days from January 1st to the corn sowing date and numOfDaysSowDate is the number of days from January 1st to the start of the R1 phase. In this test case, that would be 1.52631, since 174 days divided by 114 days equals 1.52631. Using this method, we were able to obtain a relative error of 11.6346% for our developed model and 12.54649% for the GLM.

74

I. Kralj et al.

multi plier =

num O f Days H ar vest Date − num O f Days Sow Date num O f Days R1Phase − num O f Days Sow Date

(4)

6 Conclusion and Future Work In this paper, we developed a linear machine learning model based on Generalized Linear Model, that is more accurate than Generalized Linear Model, moreover, for early prediction of corn yield with a relative error of less than 20%. We used 75% of the dataset for training, while the remaining 25% was selected for testing. Using the training dataset, we calculated weighted parameter values for both our model and GLM using data from sowing corn to harvest. After calculating the weighted parameter values, we tested the accuracy of early prediction of corn yield during the R1 phase of the corn. In the end, we succeeded in developing a more accurate linear machine learning algorithm for predicting early corn yield compared to the GLM, with an achieved relative error of 11.6346% for our model compared to 12.5464% of the GLM. In the future, we would like to improve the accuracy of early prediction of corn yield and also implement an optimization algorithm to shorten the running time of the algorithm. In addition, we will research and develop a nonlinear machine learning algorithm and compare it with the currently developed linear machine learning algorithm. A nonlinear algorithm models the dependent variable (also called the response) as a function of a combination of nonlinear parameters and one or more independent variables (called predictors). The model can be univariate (with a single response variable) or multivariate (with multiple response variables). The parameters may take the form of an exponential, trigonometric, power, or other nonlinear function. An iterative algorithm is typically used to determine the nonlinear parameter estimates, similar to the linear machine learning algorithms currently being developed. Common algorithms for fitting a nonlinear model include the Gauss-Newton algorithm, the gradient descent algorithm, the Levenberg-Marquardt algorithm, etc. Acknowledgements This work has been supported by the project IoT-field: An Ecosystem of Networked Devices and Services for IoT Solutions Applied in Agriculture funded by European Union from the European Regional Development Fund.

References 1. Moussaid, A., El Fkihi, S., Zennayi, Y., Lahlou, O., Kassou, I., Bourzeix, F., El Mansouri, L., Imani, Y.: Machine learning applied to tree crop yield prediction using field data and satellite imagery: a case study in a citrus orchard. Informatics 9(4) (2022), https://www.mdpi.com/ 2227-9709/9/4/80 2. van Klompenburg, T., Kassahun, A., Catal, C.: Crop yield prediction using machine learning: a systematic literature review. Comput. Electron. Agric. 177, 105709 (2020), https://www. sciencedirect.com/science/article/pii/S0168169920302301

Linear Machine Learning Algorithm for Early Annual …

75

3. Factors affecting crop production (2021), https://www.agriculturewale.com/factors-affectingcrop-production/ 4. Cedric, L.S., Adoni, W.Y.H., Aworka, R., Zoueu, J.T., Mutombo, F.K., Krichen, M., Kimpolo, C.L.M.: Crops yield prediction based on machine learning models: case of west African countries. Smart Agric. Technol. 2, 100049 (2022), https://www.sciencedirect.com/science/article/ pii/S2772375522000168 5. Reddy, D., Kumar, M.R.: Crop yield prediction using machine learning algorithm. In: 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS), pp. 1466–1470 (2021) 6. Manivasagam, M.A., Sumalatha, P., Likitha, A., Pravallika, V., Satish, K.V., Sreeram, S.: An efficient crop yield prediction using machine learning. Int. J. Res. Eng. Sci. Manag. 5(3), 106–111 (2022), https://journals.resaim.com/ijresm/article/view/1862 7. Cubillas, J.J., Ramos, M.I., Jurado, J.M., Feito, F.R.: A machine learning model for early prediction of crop yield, nested in a web application in the cloud: a case study in an olive grove in southern Spain. Agriculture 12(9) (2022), https://www.mdpi.com/2077-0472/12/9/1345 8. Generalized linear model | what does it mean? (2022), https://www.mygreatlearning.com/blog/ generalized-linear-models/ 9. Garden, https://garden.org/ 10. Determining corn growth stages (2020), https://www.krugerseed.com/en-us/agronomylibrary/corn-growth-stages-and-gdu-requirements.html

Agent-Based Modelling and Transportation

Multi-agent Modal Logic Evaluating Implicit Information Vladimir V. Rybakov

Abstract This paper studies relational Kripke models for computation truth values in some extended logical language. The multi-agent modal language is extended by introduction a new modal operation D(α, β). The truth of this operation is determined in a way by implicit information: it is true only if the number of states where the formula α is true is strictly less than the number of states where the formula β is true. From mathematical point of view, the paper investigates satisfiability problem for formulas in such language, we construct a mathematical algorithm verifying satisfiability the formulas.

1 Introduction Computer science and Information Sciences actively apply symbolic logic, logic may be an effective tool in research and development software. Several domains are useful here, for example it is the temporal logic (cf. [1]) [5, 6], multi-modal logics, multi-agent logics (cf. [19]) [2–4, 8–11]. For such application some special variations of non-classical logic been invented, e.g. it is the linear temporal logic LTL, which got to be very popular in computer science immediately after its introduction (cf. [8, 9, 21, 23–26]). Multi-agent logics are often used in areas of analysis obtained information, on its reliability and safety (cf. [27, 28]). Typically many instruments and approaches from logics, in their interaction, are applied, usually being combined with elements of multi-agency, parallel computing and multi-agent logics (in a sense a multi-modal logics). It seems the first substantive example of a two-modal logic is Arthur Prior’s tense logic, with two modalities, F and P, corresponding to “sometime in the future” and “sometime in the V. V. Rybakov (B) Institute of Mathematics and Fundamental Informatics, Siberian Federal University, Krasnoyarsk, Russian Federation e-mail: [email protected] Institute of Informatics Systems, Siberian Branch of RAS, Novosibirsk, Russian Federation National Research University—High School of Economics, Moscow, Russian Federation © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_7

79

80

V. V. Rybakov

past”. For example, a logic with infinitely many modalities is dynamic logic, introduced by Vaughan Pratt in 1976, it has a separate modal operator for every regular expression. For study behaviour of agents, multi-agents’ logic, modalities, are interpreted often as agent’s temporal accessibility operations, or the ones oriented to model checking, they were used widely for study interaction and autonomy, effects of cooperation (cf. e.g. [2–4, 7, 11, 22–24, 27, 28]). The area of knowledge representation also often deals with analysis of information by logical instruments (e.g. description logics) using temporal and modal logics (cf. [8–10, 13, 14, 18]). Representation of agents interaction (as a dual of common knowledge) was suggested by Rybakov [22]; using as a base agent’s knowledge (S5) modalities. Knowledge, as a concept itself, came from multi-agency, since individual knowledge may be received only from interaction of agents, learning. Multi-agent systems close to technical application and logical models for them are also very active areas in AI and CS for last decades. Various aspects including interaction and autonomy, effects of cooperation etc. were investigated (cf. e.g. [11, 27, 28]). A multi-agent logic with distances was studied and satisfiability problem for it was solved [20]. As we commented above, non-classical logics, e.g. L T L , C T L and C T L ∗ actively applied in software development (cf. [12, 15, 16]). Approach in our this paper is a new one and essentially differs from existing ones, it includes introduction of a new modal binary operation modelling implicit information (though which is not given directly in model representation, and which cannot be expressed by other modal operations). So, in this our paper we wish to touch rather novel approach to information concerning reasoning about implicit knowledge, about truth of statements, which already might be expressed by new modal logical operations. This considers a comparison, which statements are looking more close to be true comparing them to each other. We use a language of multi-modal logic extended by a new logical operation D(α, β) which compare the truth of formulas α and β on time intervals in terms the possible amount of states where they may be true. Mathematically we work with ways to invent algorithms recognizing satisfiability such modal formulas and find a deciding algorithm.

2 Denotation, Definitions, Preliminary Facts We first recall necessary for reading the paper definitions and facts concerning modal multi-agent logics. Following modern trends by a logic we understand the set of all theorems provable in a given axiomatic system, or the set of valid formulas for a certain class of Kripke frames. In particular, a normal modal logic λ is a set of modal formulas which is closed under substitution, modus ponens {α, α → β/β} and necessitation rule {α /α}, and contains all theorems of the minimal propositional modal logic K .

Multi-agent Modal Logic Evaluating Implicit Information

81

The language of modal logics consists of a countable set of distinct propositional variables p1 , . . . ; pn . . . , logical connectives of classical logic ¬, ∧, ∨, → and the unary modal operator ♦. A normal modal logic is a set of modal formulas L that contains all propositional tautologies, an axiom scheme (α → β) → (α → β), and closed with respect to substitutions, the modus ponens {α, α → β/β} and necessitation rule α/α. The minimal modal logic is denoted as the logic K. If L is a normal modal logic, then for formulas α ∈ L we write  L α or L  L α (i.e. there is a theorem of logic - it is deduced from the axioms using the postulated rules of inference). If the logic L is fixed or clear from the context, then we denote  α for simplicity. Definition 1 A frame F := F, R is a pair, where F is a non-empty set and R is a binary relation on F. The frame and its underlying set are often denoted by the same letter for simplicity. Further we consider only frames where R is a transitive and reflexive relation. Definition 2 A model is a triple M = W, R, V , where F := F, R is a frame (cf. the definition above) and V is a valuation of a set of propositional letters P in the frame F that is V : P → 2W . Dom(V ) = P is called the domain of V . A frame F = F, R is called an open subframe of frame G = G, R (denoted F G) if F ⊆ G and ∀a ∈ F ∀b ∈ G (a Rb =⇒ b ∈ F) holds. If M1 = W1 , R1 , V1 , M2 = W2 , R2 , V2 are models then we call M1 an open submodel of M2 (denoted M1 M2 ) if : 1) W1 , R1 is open subframe of W2 , R2 ; 2) Dom(V1 ) = Dom(V2 ) and ∀ p ∈ Dom(V1 ) V1 ( p) = V2 ( p) ∩ W1 . A mapping f : F, R → G, S is called p-morphism if (1) a Rb =⇒ f (a)S f (b); (2) f (x)Sz =⇒ ∃y ∈ F : f (y) = z & x Ry. We say a mapping f : M1 = W1 , R1 , V1 → M2 = W2 , R2 , V2 is a pmorphism of the model M1 into a model M2 if 1) f is a p-morphism of the frame F1 = W1 , R1 into the frame F2 = W2 , R2 ; 2) the valuations V1 , V2 are defined on the same set of propositional letters; 3) ∀ p ∈ Dom(V1 ), ∀a ∈ W1 (a |=V1 p ⇐⇒ f (a) |=V2 p). We say a frame F is an λ—frame for a logic λ if all theorems of λ are valid at F, and λ(F)—the set of all formulas valid in F—is the logic generated by F. A logic λ satisfies the finite model property (FMP) if for any α ∈ / λ there exists a finite λ-model on which α is not valid. Now we prepare denotation and definitions which we need for formulation our main results. We fix an interval partitioning I n of the set of all natural numbers N , so I n is a set of indexes and  [ci , ci1 ], ci < ci+1 . N= i∈I n

The interval linear F P-frame represents a special structure of the following form: F FP = 



i∈I n

[ci , ci1 ], ≤

82

V. V. Rybakov

or (it might be) F F P = [ci , ci1 ], ≤ or (even - for infinite time interval) F F P = [ci , ∞], ≤ , where ≤ is the standard linear order relation on natural numbers. The language of our logic is the standard language of modal propositional logic extended by a binary logical operation D(x, y), which may be applied to any (arbitrary) formulas—cf.— D(α, β). The valuations V of a set Pr op of propositional variables in frames F F P from  FP ∈ Pr op, V ( p) ⊆ i [ci , ci1 ]. models M defined as follows, ∀ p  The basis of any such model is i [ci , ci1 ]—the set of natural numbers N . If a ∈ M F P , and a ∈ V ( p) then write (M F P , a) |=V p and we say that p is true on state (element) a. Since we are interested in multi-agent logics, we will keep in the language several modal operations ♦1 , . . . , ♦n and accept several different valuations V1 , . . . , Vn of letters—each one for each supposed agent. This one will give an unusual and big distinction from standard approach and makes problems and considerations interesting. So we switch indexes for modal operations and valuations. The essence of this distinction in immediately visible in the definition of truth valuers for formulas below. Valuations of variables Pr op may be extended to formulas as follows: • Truth of Boolean connectives ¬, ∧, ∨, → is defined in the standard way; • (M F P , x) |=V j i α ⇐⇒ ∃i ∈ N : x ∈ [ci , ci+1 ] =⇒ ∀ y ∈ [ci , ci+1 ] (x ≤ y ≤ ci+1 =⇒ y |=Vi α); • (M F P , x) |=V j ♦i α ⇐⇒ ∃i ∈ N : x ∈ [ci , ci+1 ] =⇒ (∃y ∈ [ci , ci+1 ] (x ≤ y ≤ ci+1 & (M F P , y) |=Vi α)); • (M F P , x) |=V j D(α, β) ⇐⇒ ∃i ∈ N : x ∈ [ci , ci+1 ] and the number of states on the segment [ci , ci+1 ] in which the formula α is true w.r.t. valuation V j , is strictly less then the number of states of this segment at which the formula β is true. The binary logical operation D(α, β) will also be used in an unusual way. We may interpret it as an expert assessments; that is a sort of comparison statements in an implicit situation—when precise amount of states where the statements are true is not known in precise numerical value. The introduced constraint on the ♦ modal operator causes, leads, for example, to the implementation on the introduced M F P models of the following properties’: (1)  p ∧ ♦♦¬ p, and as consequence  p ∧ ♦k ¬ p, k > 1. (2) ¬ p ∧ ♦ p. (3) D(α, β) ∧ ♦¬D(α, β).

Multi-agent Modal Logic Evaluating Implicit Information

83

We also admit a big restriction in use of the language. In what follows we will consider only formulas where the operation D(α, β) may occur only once and not may occur inside subformulas of any modal operation as an internal logical operation. The reason for that is a technical one (but important) which we will comment after all main proofs. Definition 3 The logic L F P is the set of all such special formulas that are true on all multi-modal models M F P defined above.

3 Formulation of Main Results, Satisfiability Problem For some class of models K, a formula α is said to be satisfiable (in a given class K) if at some element of some model from this class this formula is true. In our case for logics L F P the class of models is given by the set of models of the form M F P . Thus, a formula is satisfiable in logic L F P if there exists a specific model M F P with some valuation V , such that, for some element x ∈ M F P of this model, the statement (M F P , x) |=V φ holds. Satisfiability is directly related to decidability: the logic is decidable if there is an algorithm that verifies, for any given formula φ, whether φ ∈ L F P is true. It is clear that φ ∈ L F P if and only if the formula ¬φ is not satisfiable. And vice versa, φ is satisfiable if and only if ¬φ ∈ / LFP. Thus, it is necessary to find an algorithm that checks the satisfiability of formulas at such finite models, where the size of models is limited by the value of some computable function from the length of formulas φ. In this our paper we use some similar ideas as we used in our recently submitted paper. But in those paper we considered only simply modal (single modal) logic but not an multi-agent logic. This makes very essential difference in proofs and makes new obtained results appreciably more strong. So let we directly start from Theorem 1 A formula f is satisfiable in a model introduced earlier (cf. M F P =   i∈I n [ci , ci1 ]) iff it is satisfiable in a special model based at a frame of kind F F P in its interval [c1 , c2 ] size of which is computable from the size of f . Proof Here as in those mentioned paper, we consider a modal formula f with propositional letters Pr op( f ) = { p1 , p2 , . . . , pk }. Assume that this formula be satisfiable at some element x of the model M F P , e.g. (M F P , x) |=V f . We need to show that in this case the formula f also holds at some finite model MM c , whose structure and size to be determined later in the course of the proof. Without loose of the generality, we can assume that x ∈ [1, c1 ]. We denote by Sub( f ) the set of all subformulas of the formula f , so Sub( f ) is closed w.r.t subformulas. Let S R := {A1 , A2 , . . . Ak } ⊆ Sub( f )} the set of all sets of all subformulas of the formula f , which are true at some elements of the interval [1, c1 ]. That is, for each subset Ai , all formulas of Ai are true on some element xi of [1, c1 ], and for any subformula of f if it is true at this element xi then it belongs to

84

V. V. Rybakov

Ai . If the initial model is finite we act as earlier in our mentioned paper, so for each such subset Ai , we choose and fix the ≤–maximal element xi with this property from the given segment [1; c1 ]. Let’s arrange all these elements xi in ≤-increasing order: ti0 < ti1 < · · · < tik . It is clear that the number of such elements is not greater than 2|Sub( f )| . Let M1 be the set of all shown states ti0 < ti1 < · · · < tik . Since that the proof may follow as before for single-agent case. But if the initial model is infinite the proof will be very different. Here is the main branching point, we need to consider stabilization points—states after which the appropriate sets Ai , will appear infinitely many times, and as soon as it appears a time, it will happen in future infinitely many times. After that we will need to collapse line on states after stabilization point in an appropriate loop. After what we do a proof for the loop part proceeding the all loops. The possible size of this our paper does not allow to include more useful details or to give all complete proof, but the main idea is to concentrate on stabilization states and to recover the appearing loop problem . Using this theorem, as a consequence we obtain Theorem 2 The problem of satisfiability for formulas in the logic L F P is decidable.

4 Conclusion In this short paper we investigate the interval modal logic, in which the action of the modal operator ♦ is limited by the boundaries of the interval. In addition, the language of modal logic is extended by the operator D(x; y), the truth of which is determined implicitly and in a sense qualitatively: it is true only if the number of states on the interval [ci ; ci+1 ] where the formula is true is strictly less than the number of states in this segment where the formula is true. That is a new look at modalities and their usage, we show that even implicit information may be analyzed via mathematical logical tools. The problem of satisfiability for formulas is solved, and as a consequence, the decidability of the logic is proved. At the same time, we confess that the area is not investigated yet in complete volume. In this paper we do not allow nested or repeated occurrences of the operation D(x; y) in formulas. Following our scheme of proof this would allow appearance of infinite loops, and it would break our proof. So, the problem with nested occurrences of formulas of sort D(x; y) (and inside each other) is yet open. Acknowledgements This work is supported by the Russian Scientific Foundation (Project No. 2321-00213) and by the Krasnoyarsk Mathematical Center and financed by the Ministry of Science and Higher Education of the Russian Federation (Agreement No. 075-02-2023-936).

Multi-agent Modal Logic Evaluating Implicit Information

85

References 1. Abadi, M., Manna, Z.: Temporal logic programming. J. Symbolic Comput. 8, 277–295 (1989) 2. Babenyshev, S., Rybakov, V.: Decidability of hybrid logic with local common knowledge based on linear temporal logic LTL. In: CiE 2008: Lecture Notes in Computer Science, 2008, vol. 5028, pp. 32–41 (2008) 3. Babenyshev, S., Rybakov, V.: Logic of discovery and knowledge: decision algorithm. In: KES (2), Lecture Notes in Computer Science, 2008, vol. 5178, pp. 711–718 (2008) 4. Babenyshev, S., Rybakov, V.: Describing evolutions of multi-agent systems. In: KES (1) 2009: Lecture Notes in Computer Science, 2009, vol. 5711, pp. 38–45 (2009) 5. Babenyshev, S., Rybakov, V.: Linear temporal logic LTL: basis for admissible rules. J. Logic Comput. 21(2), 157–177 (2011) 6. Babenyshev, S., Rybakov, V.: Unification in linear temporal logic LTL. Ann. Pure Appl. Logic 162(12), 991–1000 (2011) 7. Babenyshev, S., Rybakov, V.: Logic of plausibility for discovery in multi-agent environment deciding algorithms. In: KES (3) 2008: Lecture Notes in Computer Science, 2008, vol. 5179, pp. 210–217 (2008) 8. Baader, Franz, Sattler, Ulrike: Expressive number restrictions in description logics. J. Logic Comput. 9(3), 319–350 (1999) 9. Baader, F., Küsters, R.: Unification in a description logic with transitive closure of roles. In: Nieuwenhuis, R., Voronkov, A. (eds.) Logic for Programming, Artificial Intelligence, and Reasoning, LPAR 2001. LNCS, vol. 2250, pp. 217–232. Springer (2001) 10. Baader, F., Morawska, B.: Unification in the description logic EL. Logical Methods Comput. Sci. 6(3), 1–31 (2010) 11. Belardinelli, F., Lomuscio, A.: Interactions between knowledge and time in a first-order logic for multi-agent systems: completeness results. J. Artif. Intell. Res. 45, 1–45 (2012) 12. Clarke, E.M., Emerson, E.A.: Design and synthesis of synchronisation skeletons using branching time temporal logic (PDF). In: Logic of Programs, Proceedings of Workshop, Lecture Notes in Computer Science, vol. 131, pp. 52–71. Springer, Berlin (1981) 13. Horrocks, I., Sattler, U.: A description logic with transitive and inverse roles and role hierarchies. Description Logics (1998) 14. Horrocks, Ian, Giese, Martin, Kharlamov, Evgeny, Waaler, Arild: Using semantic technology to tame the data variety challenge. IEEE Internet Comput. 20(6), 62–66 (2016) 15. Hossain, A., Laroussinie, F.: In: Gamper, J., Pinchinat, S., Sciavicco, G. (eds.) From Quantified CTL to QBF. 26th International Symposium on Temporal Representation and Reasoning (TIME 2019). Leibniz International Proceedings in Informatics (LIPIcs), Article number 11, vol. 147, pp. 1–11. Schloss DagstuhlLeibniz-Zentrum fuer Informatik, Dagstuhl, Germany (2019) 16. Huth, M., Ryan, M.: Logic in Computer Science, 2nd edn., p. 207. Cambridge University Press (2004) 17. McLean, D., Rybakov, V.: Multi-agent temporary logic T S4U K n based at non-linear time and imitating uncertainty via agents interaction. In: Artificial Intelligence and Soft Computing, 2013, Conference Proceedings, pp. 375–384. Springer (2013) 18. Sattler, U.: Description logics for the representation of aggregated objects. In: ECAI 2000, pp. 239–243 19. Shahin, R.: Based Event Languages for the Development of Reactive Application Systems— Towards Modal Software Engineering. Research Gate (2008) 20. Rybakov, V., Babenyshev, S.: Multi-agent logic with distances based on linear temporal frames. In: Artificial Intelligence and Soft Computing, 2010: Conference Proceedings, pp. 337–344. Springer (2010) 21. Rybakov, V.V.: Non-transitive linear temporal logic and logical knowledge operations. J. Logic Comput. 26(3), 945–958 (2016) 22. Rybakov, V.V.: LTKK extended by multi-agent logic K-n with interacting agents. J. Logic Comput. 19(6), 989–1017 (2009)

86

V. V. Rybakov

23. Rybakov, V.V.: Temporal logic with overlap temporal relations generated by time states themselves. Sib. Math. Rep. 17, 923–932 (2020) 24. Rybakov, V.V.: Branching time logics with multiagent temporal accessibility relations. Sib. Math. J. 62(3), 503–510 (2021) 25. Vardi, M.: An automata-theoretic approach to linear temporal logic. In: Banff, Y. (ed.) Higher Order Workshop, pp. 238–266 (1995), Available at http://citeseer.ist.psu.edu/ vardi96automatatheoretic.html 26. Vardi, M.Y.: Reasoning about the past with two-way automata. In: Larsen, K.G., Skyum, S., Winskel, G. (eds.) ICALP, LNCS, vol. 1443, pp. 628–641. Springer (1998) 27. Wooldridge, M., Lomuscio, A.: Multi-agent VSK logic. In: Proceedings of the Seventh European Workshop on Logics in Artificial Intelligence (JELIAI-2000), 2000. Springer (2000) 28. Wooldridge, M.: An automata-theoretic approach to multi-agent planning. In: Proceedings of the First European Workshop on Multi-agent Systems (EUMAS 2003). Oxford University (2003)

Multisectoral Household Location Agent-Based Simulation for Testing Policy Decisions Simon Gorecki, Seghir Zerguini , Natalie Gaussier , and Mamadou Kaba Traore

Abstract Urban transport and housing location are key points for metropolises because they are subject of measures in the context of ecological transition and to reduce greenhouse gas emissions. A LUTI (Land Use and Transport Interaction) model’s objectives is to represent and simulate the interactions between an agglomeration’s population through its residential decisions, and transportation used to meet daily mobility needs. In this article, we propose to extend LUTI models by adding housing and workplaces as agents in order to test European political policies such as low-emission areas or the prohibition on poorly insulated housing.

1 Introduction Urban citizens are continually pushing the boundaries of urban mobility. When this tendency is combined with demographic growth, urban sprawl occurs [1]. The French government and countries of the European Union have put in place several strategies to minimize pollution to reduce greenhouse gas emissions, by international agreements (such as HORIZON EUROPE [2] or COP23 [3]), and combat global warming. A French region has funded the SIMUTEC initiative, composed of laboratories, universities, and other entities). It intends to create a platform for modeling and simulation to assist local governments in the ecological transition of their regions. S. Gorecki (B) · M. K. Traore University of Bordeaux, IMS, UMR 5218, Talence, France e-mail: [email protected] M. K. Traore e-mail: [email protected] S. Zerguini · N. Gaussier University of Bordeaux, CNRS, BSE, UMR 6060, 33600 Pessac, France e-mail: [email protected] N. Gaussier e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_8

87

88

S. Gorecki et al.

The SIMUTEC platform can simulate a variety of sectoral policies, including the pricing for transportation, new transportation infrastructure, renovating housing stock for energy efficiency, establishing a green belt, as well as scenarios of rising energy prices and population growth in the area under consideration. Moreover, SIMUTEC can analyze the energy and socio-spatial issues of the proposed policy as well as the constrained household expenses (home-work transportation, housing, etc.) for each of the simulated policies. In this article we will focus on two laws under discussion at both national and European levels, and propose modifications to standard LUTI model in order to apply them. • Low Emission Zone (LEZ). A LEZ is a territory in which an access ban has been established for certain categories of polluting vehicles which do not meet some emission standards. The main objective of the LEZ is to improve air quality and life quality in a concerned zone. This law also aims at climate action: accelerate vehicle fleet renewal. To circulate in a LEZ, an “air quality” certificate is mandatory and the most polluting vehicles (identified by a “Crit’Air” stickers 5, 4, and 3) will be prohibited from circulation. • Withdrawal Energy-intensive Housing (WEH). As part of the “Climate and Resilience” law, high energy consumption housing, called “thermal sieves”, has been prohibited for rental since January 1, 2023. The objective of this law is twofold: protect tenants against too high energy bills; reduce greenhouse gas emissions. In this article, we will first present a brief state of the art of LUTI (Land Use—Transport Interaction) simulation tool. In a second time, we will present the SIMUTEC agent-based platform in both transport part and housing location parts. This will be done in order to demonstrate what extension to LUTI model can be made on both sides to integrate new tested scenarios. Then, we will illustrate our proposition by studying a simulation ok one of the two European laws before ending with a brief conclusion.

2 State of Art Urban planners are aware of the intimate relationship between land use and transportation. The fundamental tenet of transportation analysis and forecasting is that human activities are spatially separated, which implies the necessity for transportation of people. Hence the need to design and use modeling and simulation tools that consider the interactions between citizens’ transportation and their residential choices: LUTI models. In some cases, LUTI are just mathematical models, in other cases, these models are executable and can be simulated. Thus, we can observe several uses or models, or implementations of LUTI tools in the state of the art.

Multisectoral Household Location Agent-Based Simulation for Testing …

89

In [4], Wegner et al. provide a state of art of LUTI models literature where are classified around twenty models and analyze them according to 8 criteria. (1) Their unified or composite structure, made of several subsystems;—(2) The complete or partial integration of the transport system;—(3) The theoretical foundations (models based on auctions, expected utility, equilibrium, etc.);—(4) The modeling paradigm of time and space management;—(5) Dynamic simulated;—(6) Parameterization and validation of the model;—(7) The operationality;—(8) Application of the model. Moreover, we can observe other tools in literature always focused on a single point of view, such as URBANSIM, written by Hassan et al. in [5] which presents a land-use and transportation model that can be simulated, created from an urban planning point of view and applied to the Seoul Metropolitan Area. The LUTI model MOBISIM, developed by Antoni et al. to simulate the difficulties of daily and residential mobility in an urban air space, is geography-oriented [6]. RELU-TRAN is a land-use and transportation model algorithm that Alex et al. presented in [7] with an emphasis on regional economic difficulties. Because it was created to mimic the metropolitan economy and land usage for cities like Chicago (Chicago was the study case presented in the article), this simulation is mostly oriented on the economics subject. Engineering can also be a point of view for a LUTI model. This is the case in the Jean et al. work with PIRANDELLO [8] simulation which is focused on the Paris area. Finally, sustainable development can also be an orientation of simulation because it was the subject of a French national call for projects (see PLAINSUDD in [9]) that led to the MOSTAR numeric platform applied to a city. We opted to work on a multidisciplinary approach (geography, economy, urbanism) after observing the complexity of the urban phenomenon [10] allowing to describe the mechanisms of urbanization phenomenon. These models and simulations are all unique in that they are all related to group phenomena (populations, jobs, transport, economy, etc.). This is why using the paradigm of agent-based simulation is one of the LUTI platform’s additional suggestions in comparison to the state-of-the-art. Moreover, in order to integrate our objectives of testing transport and urban policies at the level of transport and land use, we propose to extend the paradigm of the LUTI model by integrating on top of the citizen mechanism, the notion of housing as a new agent in the simulation.

3 Integrating Housing and Premises as an Agent in a Land-Use Model The SIMUTEC platform is composed of several modules. First, a LUTI model that can simulate both transport and land use mechanism ((1) and (2) in Fig. 1). SIMUTEC is also composed of an M-CLIMATE module that can calculate KPI related to energy and pollution, and M-3D, a module able to render results in a 3D interactive map. The LUTI module is composed of two main modules (see Fig. 1), namely: (1) an agentbased simulation model for household and business location, and a transportation

90

S. Gorecki et al.

Fig. 1 SIMUTEC architecture

model (2) to simulate the movement of citizens in the city through different modes of transport. The land management module (1) is an agent-based model that can simulate the location choices of households and jobs. It is based on an auction mechanism that allows to model, at different given time horizons, the competition between agents on the real estate market (existing residential, tertiary, and industrial real estate stocks) and on the land market (buildable land reserves). It indicates, by geographical area, the prices that emerge on the different real estate markets, considering the different employment and activity objectives of economic agents. The competition between the different agents (citizens of the city) is based on a variable unique to each agent: utility. The trans disciplinarily mentioned above (1) manifests itself via the mechanism of utility. (1)

(2)

(3)

(4)

(5)

Uh,d,z = α1h AC z + α2h N Oz + α3h DSd − E Bd ∗ DSd − Pzh ∗ DSd

(1)

The outcome of a calculation performed for each agent is Utility (Uh,d,z ) [10]. This function’s goal is to depict the “performance” of a workplace or the “wellbeing” of a household, or profitability of a workplace. The utility is measured in monetary terms (e) and is composed of (1) accessibility to employment (liked with the transport model), (2) notoriety of an area, due to its proximity to shops and services, (3) desired surface of the current accommodation, (4) Energy bill of the dwelling (directly related to the energy class of housing in Fig. 2) and (5) Price per square meter of the zone. Each agent (household or workplace) in the simulation has a single primary goal: to maximize his own Utility value, which depends on several parameters: h for a certain Housing, d, for a Desired surface of housing, and z, because this housing is located in a Zone. These parameters introduce the notion of housing as a full-fledged agent in the simulation which is the main proposition of this chapter. This contribution allows locating households and workplaces agents in housing and premises agents. In the same time, it allows us to attach energetic data to buildings. Based on open national statistical data (INSEE [11]), the statistical thermal capacity of the dwellings, as well as the age of the dwellings are added to the data of the new agent population. Moreover, statistics on the size of households can also be added to the simulation.

Multisectoral Household Location Agent-Based Simulation for Testing …

91

Fig. 2 Households and housing repartition mechanism

The double addition, on one hand of the integration of housing as agents, and on the other hand via the addition of national statistical data to the agents, allows the simulation to add constraints during the location choices mechanism of households or establishments in order to test the public policy scenarios. Figure 2 shows the possible distribution between household agents (5 possible household sizes) and housing agents (5 housing sizes). Each dwelling (all 5 sizes) is divided into 7 energy classes which represent 35 types of dwelling allowing to simulate a realistic representation of a city based on public and open data. The addition of dwellings and establishments as agents in the simulation allows us to go further in the scenario simulation, in particular, to allow the simulation of public policies specified in the introduction section: Withdrawal Energy-intensive Housing (classes E, F, and G). This scenario will be explained in more detail in the case study section.

4 Integrating Low Emission Zone in the Transport Model The right side of Fig. 1 illustrate the architecture of the transportation model that simulates city travel (2). This module imports the land use model output (population and places location) to simulate travels from homes to workplaces. This module is based on a 4-step model [12] which can export an accessibility score for each zone (this accessibility is used by agents through the first element of the Utility function). As we can see in Fig. 3, the transport model can be separated in several steps. These steps came from the standard 4 steps model [12].

92

S. Gorecki et al.

Fig. 3 Transport model architecture—4 steps model

The first step (1) consists in getting from the land use model the locations of each agent location (households and their corresponding housing as well as workplaces and their premises) in the city. Travers are generated as well as the location of household is a starting point, and workplace location is a destination point which leads to a transport flow matrix. The Modal split phase (2) divides the transport flow matrix into 3 different subtransport matrices; each means of transport is contained in a separated matrix as well as: the pedestrian travel matrix, cycling travel matrix, and motorized mode travel matrix. The Modal choice phase (3) is dedicated to the motorized mode matrix (F M M ). This module is the entry point of a loop dedicated to transport simulation and modal choice. FiVj P =

FiMj M

1+e

  μ CiVj P −CiTj C

(2)

In this step, a Cost is calculated for both Private Vehicle travels (C V P ) and Public Vehicle travels (C T C ). These costs are injected in the above (2) formula, and allow to split flows from the F M M matrix into two separate flow matrixes: Private Vehicle flow matrix (F V P ) and Public Vehicle flow matrix (F T C ). Once flows are generated, there are simulated on real roads network in step (4): congestion simulation, traffic, and travel time along the real roads extracted from OpenStreetMap database [13]. Step (5) update the cost of travels of private and public vehicles travels. These 3 steps (3, 4, and 5) are repeated until each simulated agent has determined its optimal means of transport. The final step (6) converts costs and flow matrixes into an accessibility value, injected in each zone of the land-use model. Our proposition in this part is to integrate the management of the Low Emission Zone (LEZ) in the transport model. As a reminder, a LEZ is à zone where an access ban has been established for certain categories of polluting vehicles. Thus, we can list 3 situations that will be affected by this measure: • people who live and work in the LEZ zone (internal traffic) • people who live in the LEZ and work outside (exchange traffic)

Multisectoral Household Location Agent-Based Simulation for Testing …

93

• people who work in the LEZ and live outside (exchange traffic) For each of these cases, the withdrawal of a part of the vehicle fleet will be mandatory, we consider that a part of people concerned will be able to get a new vehicle conforming to the standard. Otherwise, the citizens unable to get a new vehicle will be automatically loaded into public transport. The ratio of people being able to renew their vehicles depends on the state aids put in place (public policy). This is an additional parameter of the simulation. We observe two possible cases: (1) optimistic situation where 80% of the people concerned by this law will be able to renew their vehicles. Pessimistic situation (2) where 20% of the people concerned by this law will be able to renew their vehicles. The proposal of this chapter lines in the implementation of two new steps in the traditional 4 steps model [12] presented in Fig. 3. These new steps will take the following form. The addition of a user removal mechanism must take place in step (3) of Fig. 4. In this new phase, the objective is to remove forbidden flows from the motorized modes matrix F M M , in order to store them in a temporary F V I matrix. The implementation of these prohibited flows is a way to simulate the application of the Low Emission Zone (LEZ) law. Based on statistical data [11], we can estimate the number of flows made with vehicles of crit’air 5, 4, and 3 (polluting vehicle). For all these vehicles, some will be changed to less polluting vehicles (via government subsidies), others will not. They will therefore be banned, and will have to make a modal shift to public transport. These vehicles are therefore removed from the VP flows and stored in a temporary matrix F V I . In a second time, when the ransport assignment is done, step (7) reinjects F V I into the public transport flow matrix F T C . Thus corresponding to the modal shift imposed bu the regulation.

Fig. 4 New proposed architecture for transport model

94

S. Gorecki et al.

5 Study Case The SIMUTEC platform can be used in various sizes of cities. We have already applied our models and simulations on cities of 300,000 inhabitants. Any space can be compatible with our simulation. Important criteria for compatibility are (1) the area must be divided into several precise geographical zones, (2) we must have access to a database of public and (3) private transport (roads), (4) we need the number of households, (5) dwellings, (6) and employment statistics In this article, we use the “Bordeaux Urban Area” as an example (about 1 million inhabitants). As explained earlier, a city must be separated into several geographic areas. This city has been divided into 42 zones based on a minimum population and employment. Each zone as well as geographical restrictions that serve as delineations (waterways, railways, motorways, ring roads, etc.). Moreover, the areas in question can be categorized into three macro-zones such as: city center—the first 13 zones which can be described by numerous public transit options but rapidly congested roadways; Small ring— 15 zones around the city center, can be described as by a good access to public transportation, better roads, and highway access; Big ring—14 zones outside of the ring road, weak access to public transportation but more space for housing and rods less sensitive to congestion. Based on these macro-zones and public data, we first constructed a reference scenario that can be considered as “the city in its actual state”. This scenario has been developed and validated as well as KPI of simulation meets actual KPI of the city (number of people per zones, transport statistics, modal choice ratio, energy consumed by zones for housing and transport, etc.). In the study case of this paper, we are focusing on a new French law: The withdrawal of Energy-intensive Housing (WEH) policy. As part of the “Climate and Resilience” law, housing with high energy consumption, called “thermal sieves”, has been prohibited for rental since January 1, 2023. The objective of this law is twofold: to protect tenants against too high energy bills; reduce greenhouse gas emissions. Dwellings are categorized according to an energy consumption scale, from A (less than 50 kWh/m2 per year) to G (more than 450 kWh/m2 per year). The WEH law introduces a gradual ban on the rental of housing: by 2035, category G, F and E housing will be prohibited. Based on national data statistics [11], information related to the housing category in our study case can be setup in the simulation through housing agent parameters. In the city, housing labeled F, E, and G affected by the WEH measure is 41.8% in private housing and 35.1% in social housing. The measure targets only households in a rental situation, only 30% of private housing are concerned. Based on these data, we can deduce the dwellings concerned by the measure. This withdrawal of energy-intensive housing in the agglomeration will concern 12.5% will concern 12.5% of the private housing stock and 35.1% of the social stock. Thus, we can compare three scenarios: an initial state which represents the initial situation, before the implementation of the new law. A second scenario is named “optimistic” with the mechanisms of WEH activated and large state aid to help the

Multisectoral Household Location Agent-Based Simulation for Testing …

95

population involved by the laws. This scenario considers that 80% of households concerned by the law will be supported by public grants. A third scenario is named “pessimistic” where only 20% of households concerned by the laws will be supported by public grants. As previously explained, a simulation is a large number of agents drawn one after the other. Each time an agent moves, its location is changed and a counter is incremented. This counter is called the draw. A simulation has 20 million draws. At the end of the simulation, every agent has reached their maximum Utility, they won’t move anymore, and this is why we consider the simulation is over. This does not mean that in the reality of a city each person is in his or her ideal housing, the fact that each agent reaches its maximum utility is due to the random placement at the beginning. To understand the visualization of the results below, it is important to understand that capture of certain KPIs of the simulation are made every 200,000 draws: every 1% of this simulation. Figure 5 above represent the evolution of several households during the simulation execution for 2 scenarios. We can observe that at start, every agent is randomly positioned. The city center is voluntarily less filled that the other macro-zones. During the execution, we can observe that they gradually relocate to more interesting places to maximize their Utility function. We observe at the end of these simulations (and also in Fig. 6) that the favorite place for households is the small ring, contrary to the big one who’s shunned by everyone. Figure 6 illustrates 2 important KPIs between the reference scenario and the Pessimistic scenario. The first one is the number of Households: we can observe that WEH law will have an impact on the location of people in the city: migration of a part of the population from the city center to the small ring. It can be explained because some old buildings can’t be renovated due to their belonging to the historical heritage of the country. In the same way, the location at the population level (a household can contain several people) acts in the same way.

number of inhabitants

Population

evolution of the simulation (%)

Fig. 5 Evolution of the household number

96

S. Gorecki et al.

Fig. 6 Pessimistic results

Energy consumed per houshold (kWh/year)

Finally, we can observe one of the main KPIs of this European objective of reducing energy expenditure: energy consumed per inhabitant on the scale of the agglomeration. The black line represents the reference simulation when no law is setup. We can observe that the energy consumed is bigger than the two other scenarios at the end of the simulation. Moreover, we can observe that the pessimist scenario would have a better result than an optimist one. It can be explained by the fact that a dwelling renovated with state funding can go from a prohibited energy category (E, F, G) to a better one (C or D) in order to be in the legal standards, but without obtaining the performance of a new home (A or B). Thus, in the pessimistic scenario, 80% of the inhabitants of housing E, F, or G will not receive subsidies, they will have to live elsewhere, in housing often with better energy capacities. This is the reason why the general energy performance is better in a pessimistic condition.

Multisectoral Household Location Agent-Based Simulation for Testing …

97

6 Conclusion In this article, we presented new modifications to the SIMUTEC platform that were previously introduced in [14]. These modifications were made to the agent-based Land Use and Transport Interaction (LUTI) model component. These two changes represent a significant improvement over the standard LUTI model, enabling the platform to test two main regulations implemented at the European level: Low Emission Zone (LEZ) and Withdrawal Energy-intensive Housing (WEH). Both of these constraints, which are imposed on a majority of large European cities, require modifications to the mechanisms in both the agent-based land use and transport simulation parts. The complexity of cities can now be deduced from individual behaviors using agent-based modeling and simulation, revealing collective behaviors that are not possible through analytical calculation or intuition. New processes, such as remote work or pandemics, which can affect transportation and lifestyle, could be added to the platform. Finally, funding for this project will be renewed in order to scale up from simulating a city to simulating an entire region, which will raise new scientific challenges such as distributed agent-based simulations.

References 1. Newman, P., Kenworthy, J.: Sustainability and Cities: Overcoming Automobile Dependence. Island Press (1999) 2. «Horizon Europe». https://www.horizon-europe.gouv.fr/ (consulté le 5 janvier 2022) 3. «COP23». https://cop23.unfccc.int/fr (consulté le 5 janvier 2022) 4. Wegener, M.: Overview of land use transport models. J. Disc. Alg. JDA 5 (2004) 5. Hassan, M., Joo, Y., Jun, C.: A review of the development and application of UrbanSim integrated land-use and transportation model. Int. J. Urban Sci. 14 (2011). https://doi.org/10.1080/ 12265934.2010.9693686 6. Antoni, J.-P., Vuidel, G.: MobiSim: un modèle multi-agents et multi-scalaire pour simuler les mobilités urbaines, pp. 50–77 (2011) 7. Anas, A., Liu, Y.: A regional economy, land use, and transportation model (RELU-TRAN©): formulation, algorithm design, and testing. J. Reg. Sci. 47, 415–455 (août 2007). https://doi. org/10.1111/j.1467-9787.2007.00515.x 8. Delons, J., Coulombel, N., Leurent, F.: PIRANDELLO an integrated transport and land-use model for the Paris area, août 2008. Consulté le: 2 mars 2022. [En ligne]. Disponible sur: https:/ /hal.archives-ouvertes.fr/hal-00319087 9. «PLAteformes numeriques INnovantes de Simulation Urbaines pour le Developpement Durable». Agence nationale de la recherche. https://anr.fr/Project-ANR-08-VILL-0001 (consulté le 5 janvier 2022) 10. Gaussier, N., Zerguini, S., et al.: «MUST-B: a multi-agent LUTI model for systemic simulation of urban policies». Groupe de Recherche en Economie Théorique et Appliquée (GREThA) (2019) 11. «INSEE». https://www.insee.fr/fr/accueil (consulté le 2 avril 2022) 12. Ahmed, B.: The traditional four steps transportation modeling using a simplified transport network: a case study of Dhaka City, Bangladesh. Int. J. Adv. Sci. Eng. Technol. Res. 1(1), 19–40 (2012)

98

S. Gorecki et al.

13. «OpenStreetMap». OpenStreetMap France. http://www.openstreetmap.fr/ (consulté le 5 janvier 2022) 14. Gorecki, S., Zerguini, S., Gaussier, N., Traore, M.K.: Development of a transport model dedicated to an agent-based simulation of land use. In: Agents and Multi-Agent Systems: Technologies and Applications 2022, pp. 301–311. Rhodes, Greece (2022)

DaFne: Data Fusion Generator and Synthetic Data Generation for Cities Ayse Glass, Kübra Tokuç, Jörg Rainer Noennig, Ulrike Steffens, and Burak Bek

Abstract In the planning of smart cities, machine learning models can support decision-making with intelligent insights. But what data sets should training processes be based on if there is not yet a city from which to collect data, or if data is not usable due to privacy issues? Synthetic data can provide a realistic representation of conditions in the city not only for machine learning experts, but also for smart city experts. For data-affine users, there are machine learning-based methods for generating synthetic data, but these have limited accessibility to data amateurs. The Platform Data Fusion Generator (DaFne) project aims to improve the usability of data generation methods for various professions. The platform with its generic functionalities should appeal to users from all domains. This paper refers to the research on how smart-city use cases can be addressed and how complex machine learning based methods can be made accessible through the platform to urban professions. Based on results from user interviews and experiments of a smart city case study, the need for a non-generic platform feature emerges. The Use Case Explorer feature provides users with a simple interface to query pre-trained machine learning models to generate data for a specific use case.

A. Glass (B) · J. R. Noennig · B. Bek Digital City Science, HafenCity University, Henning-Voscherau-Platz 1, Hamburg 20457, Germany e-mail: [email protected] J. R. Noennig e-mail: [email protected] B. Bek e-mail: [email protected] K. Tokuç · U. Steffens Hochschule für Angewandte Wissenschaften, Berliner Tor 5, Hamburg 20099, Germany e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_9

99

100

A. Glass et al.

1 Introduction Data-driven cities are growing with data challenges. The main challenges are; security and privacy [1–5], data quality [6], accuracy and representation [3], accessibility [7, 8], validity [9, 10], scalability [8] and accuracy of representations [3], management [11] and costs. Due to challenges, when the real data is not available, not in suitable form or limited, the generation of synthetic data is a promising approach. Modelling, analysing or simulating without having the mentioned challenges has an importance for the development of the design, research and application of data-related urban design projects. Synthetic data, also known as simulated data, enables the rapid prototyping of services without real data and it is researched and applied to many scientific fields [4, 5, 8, 12]. This research aims to improve the usability of synthetic data generation with a focus on the smart cities. The Platform Data Fusion Generator (DaFne) is a multidisciplinary research project that targets the domains of artificial intelligence and smart cities. It aims to integrate different approaches for data generation on a digital platform in the form of services [13]. In addition to the infrastructural provision of services, one goal is to improve the usability of these methods through a welldesigned user interface. The functionalities can be divided into generic methods and use-case-specific functions. The generic functionalities are divided into data-related, generation-related and evaluation-related services. Synthetic data can support the development and operation of smart cities, by providing data that can be used for simulations and tests without having privacy issues [4, 8, 14]. That can assist the decision-making process and improve the smart city operations. For example, for optimizing energy consumption or improving transportation, health care, light and mobility, for making predictions about urban civil protection, hazard prediction, green space irrigation [12, 15–17]. This allows city officials, planners and researchers to identify potential issues and improve the sustainability, efficiency, and livability of urban environments. DaFne includes different generic data generation techniques as outlined in Kunert’s work [13]. One example of a use-case specific DaFne functionality is the generation of synthetic citizen movement data by utilizing an agent-based simulation of pedestrian paths. This paper takes a closer look at underscoring the need for a use-case specific approach for the smart city context besides the generic techniques.

2 Methodology This paper is investigating the effectiveness of the generic platform functionalities in addressing the problems faced by urban experts. To achieve this, an experimental research methodology is employed, which is supported by user interviews. The methodology consists of several stages. Firstly, the general platform functionalities are introduced, which have been pre-set by prior research. Secondly, the platform is

DaFne: Data Fusion Generator and Synthetic Data Generation . . .

101

characterized as a product at the layers of AIaaS. Next, a user research is conducted to understand the context of use and the problem space of the users. Ultimately, the objective of the methodology is to evaluate the ability of the platform to address the needs and requirements of urban experts, and to provide insights into potential areas for improvement. The user research was also used to create a solution space in the form of a user interface design, which is not in the scope of this paper. In order to find users, urban planners from Hamburg were contacted directly. To also find interested parties independent of domains, the project was presented at different online and offline events and exhibitions related to artificial intelligence and architecture. The presumption of the user types—(1) Domain experts for smart cities, (2) data scientists, (3) data engineers, (4) developers of generative ML models, (5) developers of evaluation methods [13]—did not represent an exclusion of other interested parties. Thus, 4 people were interviewed: Two of them are urban planner and designer, one product manager for a Data Analytics SaaS and a PhD student researching data literacy in academia. The user research helps to identify the context of use and derive user requirements on the platform. The context of use analysis was carried out based on the master-student model by conducting informal contextual interviews with the recruited users [18]. The user was treated as a master by the interviewer, who is the student, because the interviewer wants to understand in detail the user’s goals regarding synthetic data and the problems related to data in smart cities. This resulted in an output of qualitative information, which was then transformed into personas. A persona is a fictional character and detailed user model that represent archetypal users and helps to identify different user needs [19]. The interviews revealed important pain points and user needs concerning synthetic data, which are briefly summarized in Sect. 4. The following case study of the agent-based pedestrian movement simulation in Sect. 6 then compares the generic platform functionality with a use-case specific approach by presenting the experiment setups and their output. In order to understand the experiment setup, the logic behind the use case development is examined in advance in Sect. 5. Finally, the paper concludes with a summary of the conducted work and gained insights followed by further work.

3 DaFne Proceedings and Explorations The platform will provide a data interface through which users can either upload their own data or access Open Data through an interface managed by the platform. The variety of available data sets is also extendable by connecting new interfaces. The data generation then can be done by three different methods. The reproduction method enables the user to generate new rows of an existing data set for either increasing the data volume for ML tasks or to circumvent privacy restrictions that would prohibit the use of the data. The rule-based method enables the user to compose a custom data set based on defined rules for the columns. Data fusion allows multiple data sets, such as private and public data, to be combined

102

A. Glass et al.

Fig. 1 Generic platform functionalities

to increase the information value of a data set. Figure 1 shows an overview of the generic functionalities. In contrast to generic methods, use-case-specific approaches provide algorithms that are designed to solve a specific urban problem.

3.1 Characterization as Artificial Intelligence as a Service In order to understand the composition of the application, the platform is discussed at the layers of Artificial Intelligence as a Service (AIaaS). Lins et al. [3] define AIaaS as “cloud-based systems providing on-demand services to organizations and individuals to deploy, develop, train and manage AI models”. By enabling the usage of generative models, the DaFne platform primarily forms an AI Software Service. The generic methods can be regarded as Machine Learning as a Service (MLaaS) as they offer the creation and the customization of ML models. For example, the reproduction service enables the user to train the models CTGAN (Conditional Tabular Generative Adversarial Networks) and TVAE (Tabular Variational Autoencoders) [20] on their own data set with possibility for custom parameterization. In contrast, use-casespecific models like pedestrian path generation simply provide an interface to query pre-trained models, making them available as inference-as-a-service. At the same time, DaFne has the characteristics of a Platform as a Service (PaaS) with the contribution feature [21]. While the internal research team is also developing generative models, the platform architecture is designed to allow contributions from external actors to enable knowledge sharing. It is important to distinguish between use-case-specific and generic approaches in this situation as well. With the hosting of urban use cases, the project aims to create an ecosystem that can leverage advancements in smart cities. Besides providing insights with synthetic data, a product feature called Use Case Explorer could inspire urban planners with new data ideas.

DaFne: Data Fusion Generator and Synthetic Data Generation . . .

103

4 User Research Through the interviews, four personas were created. Two of them are specifically relevant to the smart city domain. User 1 is an urban planner without programming skills. He can use GIS systems and Excel at an intermediate level. He is working more on strategic aspects of urban planning in inter-governmental structures, aiming to develop ideas and strategies to increase quality of life in cities through the use of technology. This is mainly done by up-scaling already existing approaches and use cases to other places, especially in the global south. One of the biggest obstacles is the lack of citizen and movement data, as collecting it requires technological know-how from stakeholders in the respective areas. According to the user, this is a particularly big problem in the global south, where surveying and infrastructure deficiencies are more prevalent than in more developed countries. For example, projects are underway to rebuild war-damaged cities in Ukraine, where participatory data collection would provide a basis for decision-making in form of citizen mobility data. The absence of the citizens themselves is the biggest obstacle in this case. An explicit user gain would be way to access, share and extend algorithms and use cases for generating urban data, especially in abandoned areas such as war zones or informal settlements. As an example, he cited a project in South Africa in which roofs in informal settlements were identified through image recognition, since there are no maps of these regions. Extending the algorithm to include other urban elements such as streets or parks could be a great help. User 2 is an urban planner and designer working on the data collection for ML models in smart city research. The data is provided to developers of ML models to gain new insights for decision making, e.g., social service planning. She can use advanced Excel and GIS, has a very good understanding of data structures and statistical methods, and intermediate Python scripting skills. Her goal is to open new perspectives for urban planning by combining different data sets such as geospatial data and socioeconomic data. The work consists of finding different interesting and useful data sets to draw inspiration from in order to expand the scope of the research. When data is available, other problems arise that are related to the quality and usability of urban data portals. The bad interoperability and comparability of these different data sets (due to differences in e.g., quantity, units, attributes, scales, etc.) is a major barrier and often leads to loss of value of the information.

5 Use Case Design Smart cities encounter various challenges in utilizing synthetic data, which necessitate addressing the quality and validity of the produced data, integrating it with real-world data, and optimizing the generation process for efficiency. To solve these challenges, algorithms and models leverage machine learning and data mining tech-

104

A. Glass et al.

niques, domain-specific knowledge, and smart city technologies. Intelligent algorithms and models can combine generative models and data augmentation techniques to create diverse and realistic synthetic data that can be integrated with real-world data to simulate urban systems and services. The DaFne project outlines steps for developing use case logic for synthetic data in smart cities. 1. Asking problems and challenges: We identify the key issues and difficulties surrounding data in smart cities, including privacy, availability, quality, accuracy, and costs. The initial step in urban design and smart city planning involves pinpointing these problems and asking questions that lead to a better design for satisfied citizens. 2. Exploring algorithms and models for analysis, synthesis, prediction: We assess and examine algorithms and models for analyzing, synthesizing, predicting, and optimizing urban systems. These algorithms can be valuable in providing reliable information or speeding up the project, particularly in design decision processes. Alongside the problem-specific questions, we also look for or generate appropriate algorithms and models. 3. Defining learning, training, and model: We define the learning, training, and model processes for the realistic model. After the computational design of the model the next step would be identifying the learning conditions. In this step, it is also important to analyze how realistic the results are and if there are non-trustable results, that might harm the city projects in which citizen life involved. 4. Data synthesize: To have a realistic synthesize, we need trustable data and data handling. In many cases, the realistic data is not available, we must synthesize. When the synthesize process relates to city design and citizens, primary methods e.g., rule-based data generation or deep generative models might be not always applicable. The new methods and models are needed. 5. Using synthetic modeling data: We use the generated synthetic data for the simulation and prediction of urban systems like mobility, safety, resource efficiency. The applications of the data can be used also in automotive, robotics, financial and administrative services, manufacturing, security, social media etc. and some of those models also can provide the information to the designers to have better cities. Not only the use case-based usage but also the generic usage of the DaFne platform might inform and improve the design process. 6. Validation: We validate the quality and validity of the generated synthetic data with real-world data. To use in the city, accuracy and reliability of the predictions has importance. The more realistic and valuable the synthetic data, better are the results which are informing the design. The design of the cities affects our everyday life.

DaFne: Data Fusion Generator and Synthetic Data Generation . . .

105

6 Experiment: Pedestrian Path Generation Urban designers can consider agent-based models as an option, even if there is no learning function involved. An experiment was conducted for DaFne, where a synthetic data set from “Grasbrook CityScope” [22] simulation tool was used to determine if generic data reproduction methods would be suitable for the designer’s specific needs. Can insights from a simulated model of Hamburg be applicable to Berlin? In cases where data is unavailable, how can existing data sets be leveraged through the DaFne platform? However, this initial experiment was unsuccessful due to various data issues, including inconsistent location points, missing values, and an unbalanced distribution of pedestrians [12]. Thus, obtaining a new data set without modelling a new area may be necessary, and a new simulation tool is required to apply the agent-based modelling approach to different cities or neighbourhoods. Therefore, path generator agent base modelling tool is designed and developed to learn about citizen behavior in urban environments to allow urban designers and decision makers to identify potential issues and take actions to improve pedestrian safety, mobility, and accessibility. The deep reinforcement learning used to solve simulation issues to understand trends on the generated paths by artificial intelligence. First prototype and the reward system to understand behavior was explained in KES conference in Verona [12]. After that, we used a flow of possible game play to model the behavior of citizens in the real world and implemented it in Python with using real data set. The agent (citizen) is placed at a randomly defined point. The movements of the agent are defined programmatic. The novel approach here is to enhance a simple randomized walk through the simulated environment by the agent which can learn. The agent is in the simulation several times and improves its life by adjusting the flow. The outcome of actions is measured by a global score, judging the behavior. For example, actions that make an actor unhappy, unsafe, e.g. a situation with long hours at the office stuck in a loop with home, will leave a minuscule score, to be heightened by walks in the park, visits to the bar etc. The agent optimizes the score not to be achieved by an exhausting trial of all possibilities, but will use an artificial intelligence algorithms e.g. gradient descent. While the agent uses the run time program each step is recorded in an output file which will be used for further iterations. The first prototype can be played by the human agents too and seen in the presented paper [12]. For the second prototype the real data extracted from HafenCity Hamburg. Duration on the road, time spent on the location and types of locations are rated by the score system. Q-learning algorithm was added to the implementation and the tests were repeated. In a later stage a longer time will be taking to generate a robust model. The results shows that the agent can learn and deep reinforcement approach is a promising method in this use case to simulate complex behavior of the pedestrians without having complex modelling time or extreme computational power. Happiness score will be validated by scientific literature, real data from survey and interviews with the citizens living in the area. The algorithm currently runs and the trends from the outcomes will be compared with the citizen interviews.

106

A. Glass et al.

7 Conclusion The DaFne platform was introduced, and a use-case-based approach was presented, explaining the experiment conducted and the challenges identified. The pedestrian path use case can be applied to any area represented in Open Street Map, and prior ML- or data-related knowledge is not necessary. However, simulation time remains a challenge when simulating newly created areas. The conducted interviews help to allocate real users from the smart city domain to the individual platform features as well as to the different AIaaS layers. User 1 turns out to be an appropriate user for the use-case-specific functionalities as well as an interested party for the contribution feature and the PaaS characteristics. A better sharing and extension of use cases and algorithms is desired. This suggests that the development teams behind the smart city strategists are also willing to share the open source algorithms on the platform. Still, the difficulty in distinguishing synthetic from real data is also noticeable. It is debatable whether the output of image recognition algorithms for identifying rooftops represents real or synthetic data. The path generator tool can be used as an inference as a service by directly querying it with an area input. This is answering the needs of the user 1, which had complaints about the poor data situation in the global south to simulate citizen data. It’s important that the outcomes of the generated data sets are visually understandable, e.g. in the form of maps or data visualizations. User 2 is experienced with data science and can be assigned to the platform’s generic functionalities because of the paint points related to poor interoperability and inconsistent data quality. Thus, the platform can be used as MLaaS for own customization and training of generative models. The results provide initial indications of the need for a use-case specific approach. Nevertheless, the results of only two users from the smart city domain cannot be considered a representative result for a user research. Although providing pre-trained smart city models is important especially to users without data expertise, the quality requirement of the platform should be to cover different data use cases as generically as possible.

8 Future Work Further work will deal with the implementation of the other generative approaches and the development of the other platform functionalities described in Sect. 3. The priority first will be the full stack integration of the generic reproduction algorithm as it is the first completed prototype with an UI design and an architectural framework. Subsequently, user tests will be performed on the first full stack prototype to evaluate and optimize its usability. User research and interviews will always be an ongoing activity during the course of the project, as initial results are not representative for the entire development and research cycle. In addition, the quality of the

DaFne: Data Fusion Generator and Synthetic Data Generation . . .

107

reproduction data will be evaluated. In parallel, other project participants are working on approaches to optimize the generic reproduction algorithm for time series and mobility data to cover different use cases.

References 1. Raghunathan, T.E.: Synthetic data. Annual Rev. Stat. Appl. 8, 129–140 (2021) 2. Hussain, A., Lasrado, L.A., Mukkamala, R.R., Tanveer, U.: Sharing is caring-design and demonstration of a data privacy tool for interorganizational transfer of data. Procedia Comput. Sci. 181, 394–402 (2021) 3. Lim, C., Kim, K.-J., Maglio, P.P.: Smart cities with big data: reference models, challenges, and considerations. Cities 82, 86–99 (2018) 4. Papyshev, G., Yarime, M.: Exploring city digital twins as policy tools: a task-based approach to generating synthetic data on urban mobility. Data Policy 3, 16 (2021). https://doi.org/10. 1017/dap.2021.17 5. El Emam, K.: Seven ways to evaluate the utility of synthetic data. IEEE Sec. Priv. 18(4), 56–59 (2020). https://doi.org/10.1109/MSEC.2020.2992821 6. Barnaghi, P., Bermudez-Edo, M., Tönjes, R.: Challenges for quality of data in smart cities. J. Data Inform. Qual. (JDIQ) 6(2–3), 1–4 (2015) 7. Lundqvist, B.: Big data, open data, privacy regulations, intellectual property and competition law in an internet-of-things world: the issue of accessing data. In: Personal Data in Competition, Consumer Protection and Intellectual Property Law: Towards a Holistic Approach? pp. 191– 214 (2018) 8. Almirall, E., Callegaro, D., Bruins, P., Santamaría, M., Martínez, P., Cortés, U.: The use of synthetic data to solve the scalability and data availability problems in smart city digital twins (2022). arXiv preprint arXiv:2207.02953 9. Paleyes, A., Urma, R.-G., Lawrence, N.D.: Challenges in deploying machine learning: a survey of case studies. ACM Comput. Surv. 55(6), 1–29 (2022) 10. Martí, P., Serrano-Estrada, L., Nolasco-Cirugeda, A.: Social media data: challenges, opportunities and limitations in urban studies. Comput. Environ. Urban Syst. 74, 161–174 (2019) 11. Amovi´c, M., Govedarica, M., Radulovi´c, A., Jankovi´c, I.: Big data in smart city: management challenges. Appl. Sci. 11(10), 4557 (2021) 12. Glass, A.: Synthetic pedestrian routes generation: exploring mobility behavior of citizens through multi-agent reinforcement learning. Procedia Comput. Sci. 207, 3367–3375 (2022) 13. Kunert, P., Krause, T., Zukunft, O., Steffens, U.: A Platform Providing Machine Learning Algorithms for Data Generation and Fusion—An Architectural Approach. Technical report, HAW Hamburg (2022) 14. Cortés, A., et al.: Deep air—a smart city AI synthetic data digital twin solving the scalability data problems (2022) 15. Ullah, Z., Al-Turjman, F., Mostarda, L., Gagliardi, R.: Applications of artificial intelligence and machine learning in smart cities. Comput. Commun. 154, 313–323 (2020) 16. Nikitas, A., Michalakopoulou, K., Njoya, E.T., Karampatzakis, D.: Artificial intelligence, transport and the smart city: definitions and dimensions of a new mobility era. Sustainability 12(7), 2789 (2020) 17. Del Esposte, A.M., Santana, E.F., Kanashiro, L., Costa, F.M., Braghetto, K.R., Lago, N., Kon, F.: Design and evaluation of a scalable smart city software platform with large-scale simulations. Future Gener. Comput. Syst. 93, 427–441 (2019) 18. Geis, T., Polkehn, K.: Praxiswissen User Requirements: Nutzungsqualität Systematisch, Nachhaltig und Agil in die Produktentwicklung Integrieren. Aus-und Weiterbildung zum UXQB® Certified Professional for Usability and User Experience—Advanced Level “User Requirements Engineering”. dpunkt. verlag??? (2018)

108

A. Glass et al.

19. Blomkvist, S.: Persona—an overview. Retrieved 22 Nov. 2004 (2002) 20. Xu, L., Skoularidou, M., Cuesta-Infante, A., Veeramachaneni, K.: Modeling tabular data using conditional GAN. In: Advances in Neural Information Processing Systems (2019) 21. Mell, P., Grance, T., et al.: The NIST definition of cloud computing (2011) 22. López Baeza, J., Carpio-Pinedo, J., Sievert, J., Landwehr, A., Preuner, P., Borgmann, K., Avakumovi´c, M., Weissbach, A., Bruns-Berentelg, J., Noennig, J.R.: Modeling pedestrian flows: agent-based simulations of pedestrian activity for land use distributions in urban developments. Sustainability 13(16), 9268 (2021)

Application of TDABC Systems and Their Support with ABMS Approach Michal Halaška

and Roman Šperka

Abstract Business process simulations (BPS) are considered a relevant and highly applicable method of analysis. BPS allow analysis of business processes under different conditions. There are several approaches towards simulation of business processes such as discrete event simulations (DES), agent-based modelling and simulation (ABMS) or system dynamics (SD). This research focuses on ABMS approach towards simulation of business processes and its importance for simulation of costs using time-driven activity-based costing (TDABC). Using ABMS approach, active elements of the system are represented by software agents which are programmed to follow some behavioral rules and autonomously interact with each other. This is especially important for TADBC approach to simulation of costs as simulation of resources has a direct impact on activity durations which drives the costs in TDABC approach. We present a case study, where is shown, how important is a process flow and organizational perspective for simulation of process costs using TDABC approach. The required detail of business process simulation can be achieved through the use of ABMS approach. After that, we discuss other shortcomings of traditional business process simulation techniques which might have a significant impact on allocation of costs using TDABC and its simulations.

1 Introduction Cost accounting is a branch of accounting that strives to determine complete cost of production by analyzing its variable and fixed costs. Traditional costing methods do not consider the crucial role of time, which can impact expected profits and result in inefficiencies caused by bottlenecks [1]. As per Barber, Dewhurst and Pritchard M. Halaška (B) · R. Šperka Silesian University in Opava, School of Business Administration in Karvina, Univerzitni Namesti 1934, 733 40 Karvina, Czechia e-mail: [email protected] R. Šperka e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_10

109

110

M. Halaška and R. Šperka

[2], direct costs are easily linked to cost objects such as direct labor, direct expenses, equipment rental, etc. Meanwhile, indirect costs like office equipment, maintenance, utilities, etc. cannot be easily traced to cost objects. To determine the costs of each cost unit, it is crucial to allocate the total costs incurred to each individual unit. With changes in the cost structure of organizations, including an increase in overhead costs and the adoption of different technologies [3], it is necessary to examine the processes and activities involved, cost factors, and process costs through analysis of business processes at an operational level. This is especially important in today’s highly competitive markets, where organizations must utilize every advantage to succeed. It is widely acknowledged that organizations relying on older functional paradigms struggle to compete in current market conditions [4]. Contemporary cost accounting systems should be equipped to calculate various cost components, possess adaptability, and take into account the diversity and intricacy of business processes. Business process management aims to improve business processes by employing a range of techniques, such as statistical analysis, mathematical methods, queueing theory, and optimization, to meet established key performance indicators (KPIs) at the operational, tactical, and strategic levels of management. Business process simulation is a tool used to model and analyze the performance of a business process. It helps organizations understand how changes to the process, such as changes to the control flow, data flow, or organizational structure, will impact overall efficiency and costs [5]. By using simulations, organizations can make informed decisions about process improvements before implementing them in real life, leading to a more sustainable and adaptive business. Business process simulations can be conducted through three different methods [6]: (1) DES, which uses entities resources and block charts to depict entity flow and resources allocation. (2) SD, which portrays processes as “stocks”. (3) Agent-based modelling and simulation, where active elements of the modeled system are represented by software agents that follow predetermined behavioral rules and interact with each other to make decisions. DES and ABMS are bottom-up methods, while SD is a top-down approach. According to van der Aalst [7], business process is a sequence of activities which execution results in a specific outcome. TDABC assigns business’s costs to individual activities which are building components in analysis and modelling of business processes. Costs of particular activities are furthermore aggregated into the cost of the whole process. The advantage of this method compared to others is the effort of evaluating each single activity instead of evaluation based on allocation bases. This is especially important for TDABC approach and ABMS allows to simulate business process models at much lower level of abstraction. Research in this paper explores the necessity and advantages of use of ABMS approach to business process simulations for the purpose of cost allocation using TDABC. In this paper we postulated following research questions (RQ): (1) Why an ABMS approach should be used when simulating costs via TDABC approach? (2) What are the advantages and disadvantages of application of ABMS approach when simulating costs via TDABC? The remainder of this paper is as follows. Section 2 presents an introduction to TDABC approach. Section 3 is devoted to

Application of TDABC Systems and Their Support with ABMS Approach

111

ABMS approach and its advantages. The following section presents research methodology and the next section presents results of our research. Finally, we conclude and discuss our results.

2 Time-Driven Activity-Based Costing Time-driven activity-based costing was specifically introduced to solve implementation problems related to activity-based costing approach. It was specifically designed to simplify the implementation and maintenance of the activity-based costing systems [8]. TDABC provides a simplified method of identifying and reporting complex transactions through the use of time-based equations that use multiple drivers. Several researchers have shown that TDABC can produce improved cost representations and positive results. According to Everaert et al., the key lies in time estimations, where the time required for each activity is calculated [9]. TDABC only requires two parameters to estimate: the unit cost of resource supply and the time required for the resource group to perform activity [10]. While time drivers might offer greater accuracy in cost allocation, they can also be more expensive to measure. But this comes with the advantage of reducing the need for time consuming interviews and surveys, which were a hindrance in traditional activity-based costing. With TDABC, the first step of activity-based costing system implementation, defining resource pools, is eliminated, allowing for a more streamlined implementation process and quicker, more cost-effective integration with software. The second step of assigning costs remains, but time is used to directly link costs from resources to cost objects, making the design of the costing system easier and faster to implement [11]. TDABC also provides a more comprehensive accounting of business transactions by utilizing time-based equations that consider the time involved in a particular process [12]. The use of time drivers makes the system easier to maintain compared to the transaction drivers used in traditional activity-based costing systems. TDABC eliminates the need for activity pools and the use of quantity-based cost drivers, simplifying the cost allocation process [8].

3 Business Process Simulation and ABMS Approach Modelling and simulation are valuable methods of comprehending real-world systems through imitation at varying levels of abstraction. This approach has become widely accepted as a research methodology, similar to other established methods such as induction or deduction. According to Axelrod [13], simulation is highly valued due to its versatility and the diverse range of purposes it can serve, including prediction, performance evaluation, discovery, and more. These purposes are of great significance in business, as they aid in the improvement of business

112

M. Halaška and R. Šperka

processes by enabling the analysis of behaviour, evaluating decision-making strategies, reengineering existing processes, and designing new ones. Correctly designing and analysing business processes is crucial to prevent negative outcomes such as dissatisfied customers and poor performance, such as long response times or low service levels. It is essential to not only examine and understand processes before implementation, but also after to ensure they are functioning optimally. The reality that business processes within organizations are not static, but constantly adapting to meet the demands of a constantly evolving market, highlights the importance of regular analysis. Management often has to make decisions about processes without fully understanding their potential outcomes, particularly for organizations that strive for continuous improvement. According to van der Aalst [14], there are specific risks to business process simulations: instance context, process context, social context. However, a major challenge in current business simulation approaches is the accurate representation of resources, as noted by van der Aalst et al. [15] and Martin et al. [16]. People are often involved in multiple processes and allocate only a portion of their time to each process based on priorities and workload, making it difficult to model accurately. Performance is also influenced by workload and can vary accordingly, as demonstrated by studies such as Wickens et al. [17]. In addition, people often work part-time and in batches, causing work items to accumulate and be processed together. Simulation tools assume usually a stable process and organization, but if the flow of work becomes too prolonged, resources may choose to skip certain activities or additional resources may be mobilized. Another problem with business process simulation is that they treat resources as undifferentiated entities grouped into resource pools [18]. This leads to issues such as pooled resource allocation and undifferentiated performance and availability. This is a limitation as each resource has its own capabilities, performance, and availability. Moreover, many existing techniques adopt a resource allocation model based on First-In-First-Out allocation approach, while not taking into account range of possible existing resource patterns [19].

3.1 Agent-Based Modelling and Simulation ABMS approach is relatively new compared to other simulation methods, such as discrete event simulation and system dynamics. ABMS allows for more detailed representation of a system by using software agents as active elements, which follow behavioural rules and interact autonomously to make decisions. These agents can represent entities such as products, organizations, departments, people, etc. [20], in simulation of for example manufacturing, logistics, operational and management science, internet of things, cyber physical systems, and other complex heterogenous systems. In recent years, there has been a growing interest in using ABMS for business process execution. Some authors proposed an automated method to map BPMN (business process modelling notation) diagrams to beliefs-desires-intentions agents. They used a model transformation technique to convert BPMN models to

Application of TDABC Systems and Their Support with ABMS Approach

113

target models. However, it is not possible to provide a direct mapping from business languages to an agent execution platform due to significant difference in the models and levels of abstraction between them as BPMN notation does not have a formal behavioural semantics while implementation of agents requires it [21]. The ability for agents in an ABMS to change their behaviour without external control based on changes in their environment and operating conditions is known as self-organization [22]. This feature is important for business process modelling and simulation. In the digital and automated business world, sophisticated interactions between robots and humans, or robots and robots, are necessary and ABMS are well-suited to model and simulate these interactions. For ABMS to attain autonomy and self-organization, the agents must be able to coordinate their actions, learn, and more. The implementation of ABMS involves creating realistic scenarios using a group of self-governed agents, either as simple entities within the computer code or as highly intelligent objects. This approach is similar to a human’s ability to solve problems with multiple states, beliefs, trusts, decisions, actions, and responses. The most challenging aspect of ABMS is developing comprehensive and logical model to accurately represent the system being tested in the simulation. Disadvantages of ABMS Validation and verification are critical challenges in ABMS research and attract a lot of attention from researches. The difficulty in managing ABMS models arises as they become more complex. However, similar challenges have been encountered in the system dynamics approach, but it has not proven to be a substantial obstacle [23]. Another drawback of ABMS is that it requires the modeler to be familiar with object-oriented programming and a programming language, such as Java. Although this has been partially addressed through the use of graphical approaches, such as drag-and-drop techniques, specialized tools, toolkits, or development environments are still required for modelling the behaviour of typical software agents [20]. Currently, there is no established modelling notation for ABMS. It is more timeconsuming than discrete event simulation or system dynamics. This is due to the lack of a general framework to guide both academics and practitioners during the modelling and simulation process. Despite this, once the model is established, ABMS becomes very flexible and reusable [24]. The last significant barrier is the reluctance of managers to embrace new techniques. However, the growing influence of datadriven approaches and the increasing demand for informatics literacy are driving organizations to continually improve their practices. Advantages of ABMS One of the most notable advantages of ABMS is its capacity to model highly complicated systems, something that traditional business process management simulations struggle with. This is especially relevant in today’s organizations, which are becoming increasingly complex due to trends such as globalization and horizontal integration. The vast majority of organizations are characterized by uncertainty and complexity that go beyond intuition and traditional analytical methods [24]. Regarding the intricacy of simulated systems, ABMS provides the capacity

114

M. Halaška and R. Šperka

to analyze the behavior of these systems from two perspectives: macro and micro levels [23]. The macro level is well-suited for strategic and tactical decision making, while the micro level is ideal for operational decision making. Additionally, ABMS enhances the realism of simulated models through its ability to model human behavior and interactions such as communication, cooperation, and coordination. Additionally, ABMS allows for the modeling of interactions not just between agents but also between the agents and their environment. The software agents in ABMS also effectively represent entities involved in organizational processes.

4 Methodology In this section, we will be discussing the data and methodology used in our study of the loan application process, which is a complex real-life process of financial institution represented by event log [25]. Event log contains more than 13,000 cases, which are formed by 262,200 events each having 9 attributes and the data were collected between 1.10.20211 and 14.03.2012. The log contains three types of events. Each event name starts either with A, O or W. The A events are related to applications, the O events are related to offers sent to customers, and the W events are related to processing of work items of applications. The overall workflow of the process is as follows: after applying, small number of the applications is controlled for fraudulent behavior, the rest of them are controlled for completeness and supplemented by necessary information, after that the application is pre-accepted and the application is processed. Some applications are cancelled and the offer is sent to the rest of the customers and the contact with customer follows. In case that the customer accepts the offer, application is assessed and the loan is approved. In some cases, after assessment of the application, further contact with the customer might be required to complete the application. Before proceeding with process mining techniques, it was necessary to prepare the log data for analysis. The logs for the loan application process were available in XES format and it was crucial to verify that all events in the log had the necessary attributes, including case IDs, timestamps, and activities. In cases where events lacked the required attributes or were in the wrong format, modifications were made if possible. If not, those events, along with any missing values, were removed from the data. The output was the clean event log which was later used to discover process model that is contained in the log using Apromore.1 Process discovery in Apromore is based on split miner and the discovered model is represented in the form of a BPMN diagram (Business Process Model and Notation 2.0).2 Based on discovered process models time drivers are estimated using time enhancement of discovered process model. To show our case, we identified and analyzed activity ‘W_Afhandelen 1

Apromore—process mining tool. http://apromore.org/platform/tools/, last accessed 2023/01/19. OMG—Business Process Model and Notation 2.0. https://www.omg.org/spec/BPMN/2.0/AboutBPMN/, last accessed 2023/01/19.

2

Application of TDABC Systems and Their Support with ABMS Approach

115

Fig. 1 Discovered process model

Fig. 2 Extract of black oval in Fig. 1 (activity “W_Afhandelen leads” with average processing time of 16.94 min)

leads’. The retrieved process model demonstrates the difference between traditional approaches to cost allocation and TDABC. In this case, it is advisable to deploy an ABMS that considers process flow and organizational perspective in greater detail. ABMS simulation allows us to estimate or predict future process costs with greater accuracy than traditional approaches.

5 Results Figure 1 presents the discovered process model representing the loan application process in 2012.3 Figure 2 presents the extract of black oval in Fig. 1. In Fig. 2, one can see that the process model is enhanced with averages of processing and waiting times. For now, we are interested in processing times which determine duration of particular activities in loan application process. As can be seen in Fig. 2, the activity has average processing time of 16.94 min. However, if we take a closer look at activity ‘W_Afhandelen lead’, we can see that when the activity first appears in the trace, it has average processing time equal to 17.80 min, while in the case of its recurrence in the trace, the activity has average 3

Following parameters were used to discover process model: activity filter is equal to 100%, trace filter is equal to 60%, and parallelism filter is equal to 100%.

116

M. Halaška and R. Šperka

processing time equal to 13.34 min ( p AN O V A, α=5% = 0.6720). This has a significant impact on simulation of the process both from process flow as well as organizational perspective. Especially, since the activity ‘W_Afhandelen leads’ occurs in the process within observed time period 4755 times from which 763 were recurrent (15.48% of recurrent occurrences). This also affects the costs assigned to individual cases in the process. In case of activity “W_Afhandelen leads”, time equation would take following form: t E,W _ A f handelen_leads = 17.8 − 4.46x1

(1)

where x1 = 1 if event E with assigned activity name ‘W_Afhandelen leads’ occurred a second time or more in the trace. Otherwise, if it is the first occurrence of activity in the trace then x1 = 0. Equation 2 is then used for estimation of time required to process all events with assigned activity name ‘W_Afhandelen leads’: TW _A f handelen_leads =

4755  k=1

t Ek ,W _ A f handlen_leads =

4755 

17.8 − 4.46x1k

(2)

k=1

Assuming that one minute of work on a given activity costs 5 units. Then with an average processing time of 16.94 min and a number of occurrences of 4755, if we consider the average processing time of all occurrences of activity ‘W_Afhandelen lead’ without taking into account the time driver. The cost allocated to activity ‘W_Afhandelen leads’ would be 402,748.50 units over the observed time period. However, if we consider the discovered time driver and use an average processing time of 17.8 min for 3992 occurrences of activity ‘W_Afhandelen leads’ (first occurrences of activity ‘W_Afhandelen leads’) and an average processing time of 13.34 min for 763 occurrences of activity ‘W_Afhandelen leads’ (recurrent occurrences of activity ‘W_Afhandelen leads’). The cost allocated to activity would be 406,180.10 units. This means a difference of 3431.60 units in allocated costs when using ABMS approach for simulation of TDABC when considering the time drivers.

6 Conclusion The demand for precise and reliable cost information systems is forecasted to rise in the coming years, due to the increasing competitiveness and cost effectiveness of companies’ operations, changes in structures of organizations and further automation. Thus, there will be greater emphasis on simulation techniques used for simulation of business processes. In this research, we had following research questions: (1) Why an ABMS approach should be used when simulating costs via TDABC approach? (2) What are the advantages and disadvantages of application of ABMS approach when simulating costs via TDABC?

Application of TDABC Systems and Their Support with ABMS Approach

117

With regards to RQ1, we showed that it is necessary to go into lower level of abstraction to be able to properly simulate cost allocation using TDABC based on discovered time driver. In this case, it is necessary to go into more detail with regards to process flow and organizational perspective. In such case, the difference in cost allocation can be of significant difference with regards to individual activities with regards to both case level and trace level of cost allocation. When it comes to RQ2, there are other issues with traditional simulation techniques which can be addressed through sufficient amount of detail in simulation model provided by ABMS approach which can have a significant impact when allocating costs. Manual design of process models for documentation and communication purposes is a common practice, but it does not always capture all the nuances of how the process is executed in reality. These models tend to focus on the most common pathways, and may not consider exceptions or infrequent scenarios. Yet, in many cases, exceptions occur in a non-negligible percentage of instances of a process [26]. For example, these techniques may assume that resources only perform one task at a time and are always available. People are usually involved in multiple processes. People also do not work at a constant speed. Moreover, people tend to work part-time and in batches. Another problem facing business process simulation the fact that they treat resources as undifferentiated entities grouped into resource pools. Moreover, many of the existing techniques adopt a resource allocation model based on a First-In-First-Out allocation approach In many of the processes supported by information systems and other forms of automation, human resources are the limiting factor. Moreover, nowadays it is necessary to also consider a worker-robot cooperation. Acknowledgements This paper was supported by the Ministry of Education, Youth and Sports Czech Republic within the Institutional Support for Long-term Development of a Research Organization in 2023.

References 1. Zhuang, Z.-Y., Chang, S.-C.: Deciding product mix based on time-driven activity-based costing by mixed integer programming. J. Intell. Manuf. 28(4), 959–974 (2017) 2. Barber, K.D., Dewhurst, F., Pritchard, M.C.: Cost allocation for business process simulation models. Proc. Inst. Mech. Eng. B J. Eng. Manuf. 220, 695–705 (2006). https://doi.org/10.1243/ 09544054JEM379 3. Popesko, B., Papadaki, Š., Novak, P.: Cost and reimbursement analysis of selected hospital diagnoses via activity-based costing. E+M Ekonomie Manage. 18, 50–61 (2015). https://doi. org/10.15240/tul/001/2015-3-005 4. Tallon, P.P.: A process-oriented perspective on the alignment of information technology and business strategy. J. Manage. Inf. Syst. 24, 227–268 (2007). https://doi.org/10.2753/MIS07421222240308 5. van der Aalst, W.M.P.V., Nakatumba, J., Rozinat, A., Russell, N.: Business process simulation: how to get it right. In: International Handbook on Business Process Management. Springer (2009) 6. Halaška, M., Šperka, R.: Is there a need for agent-based modelling and simulation in business process management? Organizacija 51

118

M. Halaška and R. Šperka

7. van der Aalst, W.M.P.: Process mining: data science in action. Springer, Berlin (2016) 8. Ratnatunga, J., Tse, M.S.C., Balachandran, K.R.: Cost management in Sri Lanka: a case study on volume, activity and time as cost drivers. Int. J. Account. 47(3), 281–301 (2012) 9. Everaert, P., Bruggeman, W., Sarens, G., Anderson, S.R., Levant, Y.: Cost modeling in logistics using time-driven ABC: experiences from a wholesaler. Int. J. Phys. Distrib. Logist. Manag. 38(3), 172–191 (2008) 10. Bruggeman, W., Everaert, P., Anderson, S., Levant, Y.: Modeling logistics costs using time-driven ABC: a case in a distribution company (2005). https://www.researchgate.net/pub lication/24125837_Modeling_Logistics_Costs_using_Time-Driven_ABC_A_Case_in_a_D istribution_Company. Last accessed 15 Jan 2021 11. Namazi, M.: Time driven activity based costing: theory, applications and limitations. Iran. J. Manage. Stud. 9(3), 457–482 (2016) 12. Stout, D., Propri, J.M.: Implementing time-driven activity-based costing at a medium-sized electronics company. Manage. Account. Q. 12(3), 1–11 (2011) 13. Axelrod, R.: Advancing the art of simulation in the social sciences. Complexity 3, 16–22 (1997) 14. van der Aalst, W.M.P.: Business process simulation survival guide. In: vom Brocke, J., Rosemann, M. (eds.) Handbook on Business Process Management 1: Introduction, Methods, and Information Systems, pp 337–370. Springer (2015). https://doi.org/10.1007/978-3-642-451003_15 15. Van der Aalst, W.M.P., Nakatumba, J., Rozinat, A., Russell, N.: Business process simulation: how to get it right. In: International Handbook on Business Process Management. Springer (2009) 16. Martin, N., Depaire, B., Caris, A., Schepers, D.: Retrieving the resource availability calendars of a process from an event log. Inf. Syst. 88, 101463 (2020) 17. Wickens, C.D., Hollands, J.G., Banbury, S., Parasuraman, R.: Engineering Psychology and Human Performance. Psychology Press (2015). https://doi.org/10.4324/9781315665177 18. López-Pintado, O., Dumas, M.: Business process simulation with differentiated resources: does it make a difference? In: Di Ciccio, C., Dijkman, R., del Río Ortega, A., Rinderle-Ma, S. (eds.) Business Process Management, pp. 361–378. Springer International Publishing (2022). https:/ /doi.org/10.1007/978-3-031-16103-2_24 19. Russell, N., van der Aalst, W.M.P., ter Hofstede, A.H.M., Edmond, D.: Workflow resource patterns: identification, representation and tool support. In: Pastor, O., Falcão e Cunha, J. (eds.) Advanced Information Systems Engineering, pp. 216–232. Springer (2005). https://doi.org/10. 1007/11431855_16 20. Macal, C.M., North, M.J.: Tutorial on agent-based modelling and simulation. J. Simulation 4, 151–162 (2010) 21. Dam, H.K., Ghose, A., Qasim, M.: An agent-mediated platform for business processes. IJITWE 10, 43–61 (2015) 22. Boes, J., Migeon, F.: Self-organizing multi-agent systems for the control of complex systems. J. Syst. Softw. 134, 12–28 (2017) 23. Siebers, P.O., Macal, C.M., Garnett, J., Buxton, D., Pidd, M.: Discrete-event simulation is dead, long live agent-based simulation! J. Simulation 4, 204–210 (2010) 24. Alfonso Gomez-Cruz, N., Saa, I.L., Ortega Hurtado, F.F.: Agent-based simulation in management and organizational studies: a survey. Eur. J. Manag. Bus. Econ. 26, 313–328 (2017) 25. van Dongen, B.: BPI Challenge 2012. 4TU.ResearchData. Dataset (2012). https://doi.org/10. 4121/uuid:3926db30-f712-4394-aebc-75976070e91f. Last accessed 21 Jan 2021 26. Estrada-Torres, B., Carmago, M., Dumas, M., García-Bañuelos, L., Mahdy, I., Yerokhin, M.: Discovering business process simulation models in the presence of multitasking and availability constraints. Data Knowl. Eng. 134, 101897 (2021)

Assessing the Impact of Shared-Taxi Pricing on Congestion Using Agent-Based Modeling Negin Alisoltani and Mahdi Zargayouna

Abstract The low car occupancy and the great demand for automobile transportation lead to traffic congestion in many urban areas. In large-scale networks with high shareability (opportunity for sharing the trips), a successful taxi-sharing program that increases vehicle occupancy may significantly save the roadway system’s driving costs and alleviate traffic congestion. Pricing plays an essential role in this system, as the taxi provider always seeks to maximize his benefits, and the passenger prefers a cheaper fare if he is going to share his taxi. So pricing can impact the performance of a shared-mobility system and, consequently, the network traffic. In this research, we define a pricing scheme based on the shareability concept to consider the impact of trip fare on the traffic situation. To model the passengers’ and taxi providers’ behavior, we use an agent-based approach to model the taxi-sharing service. We use real-world data from the city of Lyon in France to assess the behavior of the proposed taxi-sharing system under different pricing conditions. We implement two scenarios with different maximum fares acceptable by the passengers to see the impact of pricing on congestion.

1 Introduction The significant travel demand for personal car transportation and low occupancy lead to traffic congestion, an increasingly important issue in many urban areas with rapid population and economic growth [1]. In [2], the authors show that in large-scale networks where the opportunity for sharing passengers’ trips is high, a successful ride-sharing program that increases vehicle occupancy may significantly save the roadway system’s driving costs and alleviate traffic congestion. Taxi-sharing is a type of ride-sharing where the driver is just a professional taxi driver. Recently, ridesharing is expanding from traditional private car ride-sharing to taxi-sharing [3], and

N. Alisoltani · M. Zargayouna (B) Université Gustave Eiffel, IFSTTAR, GRETTIA, Boulevard Newton, F-77447 Marne la Vallée Cedex 2 Paris, France e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_11

119

120

N. Alisoltani and M. Zargayouna

soon, to autonomous vehicle taxi-sharing [4] and big taxi providers around the world are becoming reputed for providing shared services. In a taxi-sharing system, the passenger, the ride provider, and the dispatcher are the main parts. The passenger seeks a ride to pick her/him up at the origin point and drop her/him off at the desired destination within a time interval. The ride provider has a fleet of taxis ready to serve the passengers’ requests. The dispatcher receives the requests and the fleet information and tries to find the best matches on short notice. In such a system, provider and passenger criteria determine service efficiency and impact the service capability to reduce congestion. The impact of taxi-sharing and Uber-like services on traffic congestion has been studied in many pieces of research in different urban networks and contradictory conclusions have been claimed. The authors, in [5], define the concept of shareability and show that the ability of shared services to reduce congestion highly depends on this concept. Shareability is the potential for sharing trips, and it is different for different networks with different service demand conditions. They consider both passengers’ and providers’ objectives and constraints to model the service and cluster the ride requests based on the concept of shareability. However, they do not consider pricing a necessary criterion for both passengers and providers. In the current research, we define a pricing scheme based on the shareability concept defined in [5] to consider the impact of trip fare on the traffic situation. The passengers expect to pay less when they share their ride, as their travel time would increase. So if the taxi-sharing fare is higher than a specific price, they will not be willing to use the service and may reject the offer. This behavior can have major impacts on the shareability in the system. In [6], passengers can communicate with multiple vehicles and choose the offer according to their individual preferences, such as the earliest starting time, lowest trip time, and lowest cost. So if the offered fare exceeds the lowest cost, they may reject it. In [7], the authors propose monetary constraints to model a more realistic taxi ride-sharing system. These constraints provide incentives for both passengers and taxi drivers. Passengers will not pay more compared to no ride-sharing and get compensated if their travel time is lengthened, and taxis will make money for all the rerouting distance due to ride-sharing. In this paper, we consider the maximum trip fare for the passengers. If the price of the shared taxi is higher than this maximum fare, the passenger may reject the offer. This maximum fare can depend on different parameters, such as the price of public transportation and traditional taxi services, the passenger’s profile, and priorities. Defining a pricing scheme that ensures both passengers’ and providers’ benefits is essential to consider these monetary constraints in modeling a taxi-sharing service. Various pricing schemes have been proposed for the taxi-sharing systems [8–10]. In a recent survey on taxi-sharing in [11], the authors classify the proposed pricing schemes in the literature into four categories: number of passengers-based, travel distance-based, trip urgency-based, and hybrid pricing schemes, which integrates the three previous categories. However, an important factor in this classification is missing. The time taken to cover the same distance can be different based on the traffic situation in the network. Gurumurthy et al. [12] consider a travel time-based road-pricing policy where all the major network links carry a toll based on travel time

Assessing the Impact of Shared-Taxi Pricing on Congestion Using . . .

121

on the link for all road users during the morning and evening peak periods and assess the impact of this pricing on the mode share. Xu et al. [13] presents a dynamic pricing strategy with a time-varying commission rate. Wong et al. [14] propose to impose a surcharge on taxi customers who take taxis during peak hours and/or travel towards congested areas. In our method, we propose a new pricing scheme to consider the travel distance and the number of passengers, considering the traffic situation in the taxi dispatching computations. We use an agent-based approach to model passenger and taxi provider behavior. The agent-based approach offers a way to capture both supply and demand at a microscopic level, considering individual accessibility, available choices, and personal tastes and needs [15]. Therefore, this approach can make the possibility to evaluate our taxi-sharing system from the passengers’ (transport cost, satisfaction, and service quality) and the providers’ (operational cost, incomes, and fleet configuration) point of view [16, 17]. In addition, it can easily assess the impacts on the transportation network criteria such as energy consumption and emissions, shifts between transport modes, network demand satisfaction level, and network congestion [18]. This paper uses an agent-based approach to model the taxi-sharing system, considering passengers, taxis, and dispatchers as agents. Our previous study proposed a solution for the dynamic traffic conditions for a real-time ride-sharing service [19]. We use the same approach in this paper to consider large-scale network traffic. We define two models to deal with dynamic traffic conditions: the plant model and the prediction model. The current mean speed in the network will be used over the next 10 min to predict travel times for the dispatcher’s calculations. Then, taxis and personal vehicle travel are simulated. The traffic situation is updated every 10 s using a trip-based MFD model as the plant model to represent the traffic dynamics. The remainder of the paper is organized as follows. First, Sect. 2 proposes a multi-agent model for the taxi-sharing model. Then Sect. 3 presents a pricing scheme for this system. Section 4 explains how we solve the dispatching problem. Section 5 presents the numerical experiment, and finally, Sect. 6 concludes the paper.

2 Multi-agent System for the Shared Taxi Service The multi-agent system designed for the taxi-sharing service is shown in Fig. 1. The main components of this model are the passenger agents, the taxi agents, and the dispatching system. Each component is described as follows. The passenger sends her/his request for a trip, defining the number of demanded seats, the time window (earliest pick-up time and latest drop-off time) Passenger agent: The passenger sends his request for a trip via an application. This request contains the origin location, the destination point, the desired time window (earliest pick-up time and latest arrival time), and the number of demanded seats. Then he receives the dispatcher’s choices, with different prices and waiting/travel times.

122

N. Alisoltani and M. Zargayouna

Fig. 1 Multi-agent system for taxi-sharing

The passenger agent can choose one of the possible options, considering his behavioral rules and constraints. These constraints include the maximum fare he is willing to pay and the maximum number of sharing. The number of sharing is a concept presented by [20], and it shows the maximum number of other passengers on board simultaneously with the current passenger. Taxi agent: The taxi provider has a fleet of taxis, and the dispatcher has access to the fleet information, including the number of available taxis, the allowed stop locations, and the capacity of taxis at any time. The taxis are connected to the provider. They can have two situations, whether moving in the network to serve the onboard passengers or waiting in the allowed stop locations to be assigned to the new passengers. The dispatcher has access to the current location and the number of onboard passengers for each taxi at any time. Dispatcher: As mentioned before, the dispatcher has access to the taxi fleet information. When it receives the new trip request, it calculates the best offer for the passenger and sends the choice(s) to the passenger. The passenger will choose one of the options (if more than one) and return his response to the dispatcher. Then the dispatcher updates the schedule of the selected taxi and sends the new schedule to the taxi. This component solves an optimization problem in real-time to find the best offers for the passengers. The performance of this solution method is explained in Sect. 4.

3 Taxi Pricing Scheme The taxi price calculation is shown in Eq. 1 where PFi xed is the fixed price that each passenger should pay at the beginning, Pdistance is the price per distance, and Dt is the total travel distance for taxi t. Price = PFi xed + Pdistance × Dt

(1)

When every two passengers share a taxi trip, we can have two different situations for the ride [5]. In the first situation, which we call “shareable-FIFO”, the taxi has to

Assessing the Impact of Shared-Taxi Pricing on Congestion Using . . .

123

Fig. 2 The number of onboard passengers for different trip-sharing situations

pick up the first passenger and then the second passenger, dropping them off at the destination point considering their pick-up sequence. In the second situation, which we call “shareable-FILO”, the taxi drops off the second passenger before the first passenger (Fig. 2). We propose a pricing scheme based on these sharing situations. 1. Shareable-FIFO: The total travel distance for the taxi is Dt = d1 + d2 + d3 , where d1 is the distance from the first origin to the second origin, d2 is the distance from the second origin to the first destination and d3 is the distance from the first destination to the second destination. For the first passenger, the travel distance is d1 + d2 , and for the second passenger, it is d2 + d3 . In this case, the fare for each passenger can be computed as follows: f ar e1 = PFi xed /2 + Pdistance × (d1 + d2 /2)

(2)

f ar e2 = PFi xed /2 + Pdistance × (d2 /2 + d3 )

(3)

2. Shareable-FILO: if we show the distance of each link by d1 , d2 and d3 , the total travel distance is d1 + d2 + d3 for the taxi, d1 + d2 + d3 for the first passenger and d2 for the second passenger. So for the second passenger, the travel distance is the same as without sharing. However, he/she is on board with another passenger. So to ensure equity, this passenger has to pay half of the fixed price, and the fare for each passenger is computed as below: f ar e1 = PFi xed /2 + Pdistance × (d1 + d3 )

(4)

f ar e2 = PFi xed /2 + Pdistance × (d2 )

(5)

It is important to mention that the shareable-FILO situation happens when the two origins or destinations are very close, and the system can serve the first passenger within her/his time window. A taxi trip, which starts from a stop location and ends

124

N. Alisoltani and M. Zargayouna

at a stop location, can contain different combinations of these two situations. From Eqs. 2–5, we can conclude that for any combination of the sharing situation, the trip fare for passenger k can be computed as: k × dorgk desk + f ar ek = PFi xed /Nt + Pdistance × (xorg k desk



xikj × di j

i j∈L ,i, j=orgk ,desk

No

ij

)

(6) where Nt is the total number of passengers for taxi t, di j is the distance for the link between each two taxi stops i and j (to pick up or drop-off the passengers), xikj is a binary variable which is equal to 1 if the passenger k is on board from point i to point j, L is the set of links, orgk and desk are the origin and destination points for ij passenger k, and No is the number of onboard passengers when the taxi is taking link i j.

4 Dispatching Algorithm Using the pricing scheme presented in the previous section, we can compute the trip fare for each passenger. Each passenger agent can accept or reject an offer, considering her/his threshold for the shared taxi fare. Once the dispatcher receives the requests, it starts to find the best offer for the passengers that respects all their constraints, including the time window constraint, the maximum waiting time, the maximum number of sharing, and the maximum trip fare. Also, it has to ensure the taxi capacity constraint. The best schedule is the one that maximizes the revenue for the taxi provider and maximizes the number of assigned passengers. The dispatcher in our model uses the algorithm in Algorithm 1 to find the best offers. It has two main parts, to compute the new routes for taxis that are waiting at the stop locations and also for the moving taxis that are already serving passengers. For each group of received requests, it starts creating branches of in-sequence pick-up and drop-off points. For each point that will be added to the route, the dispatcher has to ensure that all the constraints are respected. Then when it finds the feasible branches, it puts them in a set called O f f er s. In the end, the dispatcher sends the optimal offer(s) to the passenger, and if the passenger accepts it, the dispatcher updates the taxi schedule and sends it to the taxi.

5 Experiments The dispatcher uses estimates for the predicted travel time obtained from the “prediction model” [21]. When the rides are executed, a gap usually exists between the estimation and the real traffic condition. The “plant model” represents the real traffic condition. Distinguishing the prediction and the plant models can provide a realistic

Assessing the Impact of Shared-Taxi Pricing on Congestion Using . . .

125

assessment of the system’s functioning. We use real data from the Lyon network in our simulations. The prediction model is based on the last observed travel times. In contrast, the considered plant model is a trip-based Macroscopic Fundamental Diagram (MFD) model able to reproduce the evolution over time of mean traffic conditions for a full road network using the MFD as a global behavioral curve. The macroscopic fundamental diagram (MFD) overviews the network states [22, 23]. The origins/destination set contains 11,314 points. The network is loaded with travelers of all ODs with a given departure time representing the morning peak hour from 6 to 10 AM. The number of trips during this period is 484,690. Algorithm 1: Dispatching algorithm input: New requests, travel distances (di, j ), travel times (tti, j ) set of taxis (T ), set of origin points ( P ), set of destination points( A), time windows ( E Pi , L Di ), number of seats demanded (sk ) for passenger k , number of sharing (n ishar e ), taxi capacity (Cap), weights of objective function (α , β ) output: Taxi schedules for origin p ∈ P do for c-schedule, the schedule of taxi t ∈ eT do if Detour is possible from any of the remaining origins on c-schedule then Build the r e-schedule by adding the p after origin ; if p is feasible for time window, capacity, and number of sharing constraints on c-schedule then if des the destination of p is feasible for time window on c-schedule then Create new schedule n -schedule by adding p and des to c-schedule; Put n -schedule to the O f f er s set;

Find the optimized route optimal -schedule∈ O f f er s ; while Not all the points in A are assigned do Closest waiting taxis t ∈ T ; Create initial routes set S from remaining origins in origin set P ; while S is not empty do Find the optimized route s ∈ S (in terms of objective function); Find the set of points S P that can be added to s ; for sp ∈ S P do if sp is feasible for time constraints on s then Compute new vehicle capacity ; if sp is feasible for capacity and number of sharing on s then Create new route ns by adding the point sp to the route s ; Add route ns to the routes set S ; if All sp ∈ S P are non-feasible in route s then if number of route origins = number of route destinations then Put route s in the offer set O f f er s ; else Remove route s from routes set S ; Find the optimized route optimal -r oute ∈ O f f er s that maximizes the taxi price; Compute f ar ek for the optimal solutions in O f f er Send the pick-up and drop-off time of the optimal solution(s) with f ar ek for each solution to passenger k if Passenger k accepts the offer then Assign the optimal -r oute for this offer to the taxi t ; Remove pick-up points on optimal -r oute from P ; if t is a waiting taxi then Remove t from waiting taxi sets T and add it to the en-route vehicles set eT ;

126

N. Alisoltani and M. Zargayouna

Among these trips, the origin and destination points of 205,308 trips are inside the studied network. We assume that the demand for taxis is 40% of these trips. The taxi provider has 1000 taxis. To estimate the prices and taxi operators in Lyon, we use the information in [24]. If the passenger accepts the taxi provider’s offer, the trip will be made by a taxi-sharing service. Otherwise, we consider that the passenger uses a personal car (or a traditional taxi with the same functionality) to make the trip.

5.1 Results We assess two scenarios. First, we set the maximum fare max kf ar e for each passenger agent equal to the taxi fare without sharing. So if the offered fare is more than this, the agent may reject the offer. In the second scenario, we assume that the passengers expect a cheaper fare equal to 75% of the fare in the first scenario and see the impact of the maximum fare on the taxi-sharing system in terms of changing traffic conditions. The traffic situation for these two scenarios is shown in Fig. 3 compared to two other situations. “No service scenario” shows the situation when all the demand in the network is made by private vehicles. The traditional taxi service situation is when no passenger shares his ride. In this case, the taxi travel distance will increase, and the network will face more traffic. Our proposed taxi-sharing system can significantly reduce traffic congestion. However, this reduction highly depends on the passengers’ preferences. This reduction would be less when the passengers have more strict limitations on the maximum taxi fare. Table 1 shows the simulation results for the taxi trips. As shown, when the passengers desire to pay a price that is less than 75% of the taxi price for a shared ride, the rejection rate is 3.52% more than when they accept to increase their desired maximum fare to be less than the taxi price. However, in both situations, the rejection rate is low as the pricing scheme and taxi-sharing system presented in this research can find the best fares to respect both passengers’ and providers’ expectations.

Fig. 3 Traffic situation comparison

Assessing the Impact of Shared-Taxi Pricing on Congestion Using . . . Table 1 Simulation results for two pricing scenarios Simulation Rejected offers (%) Total travel distance (km) Scenario 1 Scenario 2

0.78 4.30

434,004.8 563,385.9

127

Average passenger waiting time (s) 76.46 49.36

6 Conclusion In this research, we defined a pricing scheme based on the shareability concept [5] to consider the impact of trip fare on the traffic situation. To model the passengers’ and taxi providers’ behavior, we used an agent-based approach to model the taxi-sharing service. We used real data from the city of Lyon in France to assess the behavior of the proposed taxi-sharing system under two scenarios with different maximum fares acceptable by the passengers to see the impact of pricing on congestion. The results show that a reasonable pricing scheme for the taxi-sharing system that can ensure both passengers and taxi providers can help reduce congestion in the network. In this research, we assessed the impact of pricing on network traffic. However, the trip fare is the same during different hours of the day. In future research, we will consider the impact of traffic on time-dependent pricing.

References 1. Pisarski, A.: Commuting in America III: The Third National Report on Commuting Patterns and Trends, vol. 550. Transportation Research Board (2006) 2. Alisoltani, N., Leclercq, L., Zargayouna, M.: Can dynamic ride-sharing reduce traffic congestion? Transp. Res. Part B Methodol. 145, 212–246 (2021) 3. Martinez, L.M., Correia, G.H.A., Viegas, J.M.: An agent-based simulation model to assess the impacts of introducing a shared-taxi system: an application to Lisbon (Portugal). J. Adv. Transp. 49(3), 475–495 (2015) 4. Krueger, R., Rashidi, T.H., Rose, J.M.: Preferences for shared autonomous vehicles. Transp. Res. Part C Emerg. Technol. 69, 343–355 (2016) 5. Alisoltani, N., et al.: Space-time clustering-based method to optimize share ability in real-time ride-sharing. Plos One 17(1), e0262499 (2022) 6. Goel, P., Kulik, L., Ramamohanarao, K.: Optimal pick up point selection for effective ride sharing. IEEE Trans. Big Data 3(2), 154–168 (2016) 7. Ma, S., Zheng, Y., Wolfson, O.: Real-time city-scale taxi ridesharing. IEEE Trans. Knowl. Data Eng. 27(7), 1782–1795 (2014) 8. Agarwal, S., et al.: The impact of ride-hail surge factors on taxi bookings. Transp. Res. Part C Emerg. Technol. 136, 103508 (2022) 9. Nourinejad, M., Roorda, M.J.: Agent based model for dynamic ridesharing. Transp. Res. Part C Emerg. Technol. 64, 117–132 (2016) 10. Danaf, M., Abou-Zeid, M., Kaysi, I.: Modeling travel choices of students at a private, urban university: insights and policy implications. Case Stud. Transport Policy 2(3), 142–152 (2014) 11. Qiu, J., Huang, K., Hawkins, J.: The taxi sharing practices: matching, routing and pricing methods. Multimodal Transp. 1(1), 100003 (2022)

128

N. Alisoltani and M. Zargayouna

12. Gurumurthy, K.M., et al.: Benefits and costs of ride-sharing in shared automated vehicles across Austin, Texas: opportunities for congestion pricing. Transp. Res. Rec. 2673(6), 548–556 (2019) 13. Xu, K., Saberi, M., Liu, W.: Dynamic pricing and penalty strategies in a coupled market with ridesourcing service and taxi considering time-dependent order cancellation behaviour. Transp. Res. Part C Emerg. Technol. 138, 103621 (2022) 14. Wong, R.C.P., Szeto, W.Y.: The effects of peak hour and congested area taxi surcharges on customers’ travel decisions: empirical evidence and policy implications. Transport Policy 121, 78–89 (2022) 15. Zargayouna, M., Balbo, F., Scemama, G.: A multi-agent approach for the dynamic VRPTW. In: Proceedings of the International Workshop on Engineering Societies in the Agents World (ESAW 2008), Saint-Etienne (2008) 16. Zargayouna, M., et al.: Multiagent simulation of real-time passenger information on transit networks. IEEE Intell. Transp. Syst. Mag. 12(2), 50–63 (2020) 17. Zargayouna, M., et al.: Impact of travelers information level on disturbed transit networks: a multiagent simulation. In: 2015 IEEE 18th International Conference on Intelligent Transportation Systems (ITSC), pp. 2889–2894. IEEE (2015) 18. Alisoltani, N., Zargayouna, M., Leclercq, L.: Real-time autonomous taxi service: an agentbased simulation. In: Agents and Multi-Agent Systems: Technologies and Applications, vol. 2020, pp. 199–207 . Springer, Singapore (2020) 19. Alisoltani, N., Zargayouna, M., Leclercq, L.: A multi-agent system for real-time ride sharing in congested networks. In: Agents and Multi-agent Systems: Technologies and Applications, vol. 2020, pp. 333–342. Springer. Singapore (2019) 20. Alisoltani, N., et al.: Optimal fleet management for real-time ride-sharing service considering network congestion. No. 19-04863 (2019) 21. Dehkordi, N.A.: Simulation-Based Optimization Frameworks for Dynamic Ride-Sharing (Méthodes d’optimisation basée sur la simulation pour le covoiturage dynamique). Doctoral dissertation, University of Lyon, France (2020) 22. Ameli, M., Alisoltani, N., Leclercq, L.: Lyon North realistic trip data set during the morning peak (2021) 23. Ameli, M., et al.: Departure time choice models in urban transportation systems based on mean field games. Transp. Sci. 56(6), 1483–1504 (2022) 24. Lyon taxi information. https://fr.statista.com/statistiques/564574/nombre-taxis-lyoncomparaison-par-arrondissement-france/

Distributed, Classical and Flexible Job Shop Scheduling Problem with Transportation Times: A State-of-the-Art Bilel Marzouki, Olfa Belkahla Driss, and Khaled Ghedira

Abstract Today, several companies have successfully distributed their planning machines in different locations or factories, which facilitates the execution of tasks, the rapidity with which tasks are performed, and decreases the delay time. The Distributed Job shop Scheduling Problem with Transportation time (DJSPT) is one of the scheduling problems, where each operation must be processed on one or different machines and its processing time depends on the used machine, and this is in a set of geographically distributed factories. Each factory contains m machines, on which n jobs must be processed. The transport of jobs between machines is made by one or several transport robots. The DJSPT, with two versions; centralized and no-centralized; combines three Np-Hard problems: The assignment problem of jobs to machines, the distribution problem of jobs to factories, and the robot routing problem. In this paper, we present the state of the art of job shop scheduling problem with transportation times in three versions: classical, flexible, and distributed according to different criteria such as the used approach, the number of transport robots (SR to single/MR to transport robots), the flexibility of job shop environment (Yes or No) and the optimization criterion.

B. Marzouki (B) ESPRIT School of Business, Tunis, Tunisia e-mail: [email protected] O. B. Driss Université de la Manouba, ESCT, Campus universitaire, Manouba 2010, Tunisia K. Ghedira Université Ibn khaldoun, Manouba, Tunisia B. Marzouki · O. B. Driss Université de la Manouba, ENSI, LARIA UR22ES01, Campus universitaire, Manouba 2010, Tunisia © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_12

129

130

B. Marzouki et al.

1 Introduction Today, several companies have succeeded to distribute their scheduling machines in different locations or factories, which facilitates the execution of the tasks, the speed of completing the execution of the tasks, and decreases the delay time. The Distributed Job shop Scheduling Problem (DJSP) is considered among the most studied scheduling problems in the literature due to its importance. This type of scheduling can be found in several areas (textile factory, automobile factories, cable assembly factories ). In the DJSP, there is a set of distributed factories composed of m machines, n jobs, and a distance between each job and each factory. The Distributed scheduling problems with transportation time and more specifically the Distributed Job shop Scheduling Problem with Transportation time with a single robot (DJSPT-SR) and many transport robots (DJSPT-MR) where the jobs are transported between machines by one or more transport robots are much more complicated than standard problems. The Distributed Job shop Scheduling Problem with transportation Time (DJSPT) combines three NP-Hard Problems: 1. The assignment problem of jobs to machines [1]. 2. The problem of distribution of jobs in different factories [2]. 3. Robot routing problem or pickup and delivery problem [3]. Therefore, the DJSPT is much more complex than standard problems. The majority of existing works are limited to the standard job shop problem with transportation time or distributed and flexible job shop scheduling problem without transportation robots. The Distributed Job shop Scheduling Problem with Transportation robots is composed of a set of factories (l = 1, 2, . . ., F ), a set of jobs (i = 1, 2, . . ., N ) and they are geographically distributed with a travel time (T Tli ). Each factory has a set of machines (Ml ) and each job i is composed of a set of Ni operations (Oi j , i = 1, 2, . . ., N; j = 1, 2, . . ., Ni ). Each job Ji is composed by (n i−1 ) transport operations (Ti,1 , Ti,2 , . . ., Ti,ni−1 ) to be made by a single robot r or by a set of r robots (r = r1 , r2 , . . ., Rl ) between two machines for each transport operation Ti, j , we find two types of movements: full transport operation and empty transport operation. Our objective is : • Affect the jobs in factories. • Determine the scheduling of jobs in each factory. In DJSPT, we chose the following hypotheses and constraints: • • • •

Each machine executes at most one operation at each time. Precedence constraint between the operations of the same job. The execution of the operation cannot be divided into several steps. Once a job is allocated to a factory, all of its operations will be processed in that factory. • Each operation is transported by a single transport robot. • Each robot must move at most one operation at each time.

Distributed, Classical and Flexible Job Shop Scheduling Problem . . .

131

2 State-of-the-Art Review Few works of the literature dealt with the classical, flexible and distributed Job shop scheduling with Transportation robots. In the next part, we cite the studies of the literature that dealt with the classical and flexible job shop scheduling problem with transportation time since we are interested in the consideration of robots for the transport of jobs between machines.

2.1 Exact Methods We start with the exact methods: [4] proposed a formulation using mixed linear programming for job shop scheduling problem with transportation time with many transport robots to minimize the makespan. Pundit and Palekar [5] developed an algorithm of branch and bound for JSPT-MR to minimize the makespan. Deliktas et al. [6] proposed novel mathematical models to single and bi-objective functions for the flexible job shop scheduling problem in a cellular manufacturing environment by taking into consideration exceptional parts, intercellular moves, intercellular transportation time, sequence-dependent family setup times, and recirculation. The authors used the scalarization method, the weighted sum method, -constraint method, and conic scalarization method. Ham [7] proposed a novel application of constraint programming for the job shop scheduling problem with transportation time using instances in the literature and proving the optimality of the instances of Bilge and ulusoy, and he proposed also a new benchmark instance.

2.2 Approximate Methods The approximate methods such as heuristics and metaheuristics have also been used to solve the classical and flexible job shop scheduling problem with transportation time due to its ability to solve these problems in a reasonable resolution time. Anwar and Nagi [8] treated the multi-objective job shop scheduling problem with transportation time with many transport robots using a forward propagation heuristic to optimize the makespan, Work in process and inventory holding costs. Authors [9, 10] proposed an approach based on tabu search to job shop scheduling problem with a single robot to minimize the maximum completion time (Cmax). Tamer et al. [11] proposed a hybrid approach based on genetic algorithm with a heuristic to solve the job shop scheduling problem with transportation time with many transport robots (JSPT-MR) to minimize the maximum completion time and they tested their approach on instances of Bilge and ulusoy. Deroussi and Norre [12] proposed to solve the flexible job shop problem with transportation time with many transport robots approach based on a simulated annealing metaheuristic with a phase of local search

132

B. Marzouki et al.

to minimize the makespan and buffer management. An hybrid approach proposed by Zhang et al. [13] based on combination of genetic algorithm and tabu search to solve the flexible job shop scheduling problem with transportation time with a single and many transport robots to minimize the makespan and the Buffer management. Gondran et al. [14] proposed a new evaluation function using Time-Lag Heuristic (TLH) to solve Job-shop Scheduling Problem with Routing (JSPR), where a fleet of vehicles transports jobs between machines in order to minimize the makespan in the first time, then maximize the quality of service. Lacomme et al. [15] addressed the job shop scheduling problem with transportation time with several transport robots. The objective is to determine a schedule of machine and transport operations with minimal makespan. They modeled the problem by a disjunctive graph and they used a local search algorithm to solve this problem. The authors used the instances of Bilge and ulusoy, Hurink and Knust and Deroussi et al. El Khoukhi et al. [16] chose to study the Job Shop scheduling Problem with transportation time (JSPT) including new additional constraints: the number of robots and their multiple transfer capacities, such as the limited capacity of machines. They proposed an approach based on the ant colony metaheuristic to solve the JSPT. Karimi et al. [17] studied the FJSPT and they proposed the imperialist competitive algorithm hybridized with simulated annealing to minimize the makespan. Bekkar et al. [18] proposed two greedy heuristics based on an iterated insertion technique to solve the flexible job shop scheduling problem with transportation time. The algorithms are started with a greedy construction method, and then an iterative destroy and recreate algorithm. The tests are made on instances of Trentesaux et al. Xu et al. [19] proposed a new dynamic algorithm based on the simulation approach and multi-objective optimization to solve the Flexible Job shop Scheduling Problem with transportation assignment to minimize the makespan, the robot travel distance, the time difference with due date and critical waiting time. Marzouki et al. [20] proposed an imporoved chemical reaction optimization metaheuristic to solve the distributed and flexible job shop scheduling problem in order to minimize the maximum completion time. The experiments are made on instances of Giovanni and Pezzela, Brandimarte and Fattahi et al. [21] proposed an improved differential evolution simulated annealing method. Simulated annealing is integrated to local search the best Pareto solutions. The greedy idea is also used to select the offspring. Du et al. [22] used a hybrid algorithm of estimation of distribution algorithm (EDA) and Variable Neighborhood Search (VNS) to solve the distributed flexible job shop scheduling problem with crane transportations (DFJSPC) in order or minimize the makespan and total energy consumption. Fontes et al. [23] proposed a hybrid PSO and simulated annealing to solve the job shop scheduling problem with transport resources and they tested the proposed approach on 73 instances from litterature. Du et al. [24] proposed a deep Q-network (DQN) model to solve a multi-objective FJSP with crane transportation and setup times in order to optimize the makespan and total energy consumption. Jiang et al. [25] proposed a mathematical model of the energy-conscious FJSP with transportation time and deterioration effect is built and the experiments demonstrate that the proposed algorithm is effective for the consider problem. Chaudhry et al. [26] have integrated both the scheduling of machines and

Distributed, Classical and Flexible Job Shop Scheduling Problem . . .

133

automated guided vehicles in the flexible job shop scheduling problem using genetic algorithms.

2.3 Distributed Methods Distributed methods are also been used to solve the classical and flexible job shop scheduling problem with transportation time. Fontes et al. [23] proposed a multi agent model to solve the job shop scheduling problem with transportation time with many transport robots (JSPT-MR) to minimize the makespan. Another work based on multi agent model proposed by Du et al. [24], they used The Russian Theory of Inventive Problem Solving (TRIZ) for process planning and mobile robot control in a production environment. Jiang et al. [25] proposed a hybrid metaheuristic approach based on the combination of genetic algorithm and tabu search in a holonic multiagent model for the job shop problem with transportation time with many transport robots (JSPT-MR), a scheduler agent applies a Neighborhood-based Genetic Algorithm for a global exploration of the search space, then a tabu search phase is used to find new and improved scheduling solutions. Chaudhry et al. [26] also proposed to solve the flexible job shop scheduling problem with transportation time with many transport robots (FJSPT-MR) using a multi agent model based on the combination of genetic algorithm with tabu search metaheuristic to minimize the makespan. Erol et al. [27] proposed a holonic multi-agent model based on the combination of genetic algorithm and tabu search to flexible job shop environment problem with a single transport robot to minimize the maximum completion time (Cmax). The authors used the instances of Bilge and Ulusoy in their experiments. Another work based on multi agent model proposed by Petrovi’c et al. [28], they used The Russian Theory of Inventive Problem Solving (TRIZ) for process planning and mobile robot control in a production environment. Nouri et al. [29] proposed a hybrid metaheuristic approach based on the combination of genetic algorithm and tabu search in a holonic multiagent model for the job shop problem with transportation time with many transport robots (JSPT-MR), a scheduler agent applies a Neighborhood-based Genetic Algorithm for a global exploration of the search space, then a tabu search phase is used to find new and improved scheduling solutions. The authors used the instances of Bilge and Ulusoy and Hurink and Knust in their experiments. Nouri et al. [30] also proposed to solve the flexible job shop scheduling problem with transportation time with many transport robots (FJSPT-MR) using a multi agent model based on the combination of genetic algorithm with tabu search metaheuristic to minimize the makespan. The authors used the instances of Bilge and Ulusoy and Deroussi and Norre in their experiment tests. Nouri et al. [31] proposed a holonic multi-agent model based on the combination of genetic algorithm and tabu search to flexible job shop environment problem with a single transport robot to minimize the maximum completion time (Cmax).

134

B. Marzouki et al.

3 Analysis and Discussion We organize the different studies of the literature dealing with the classical and flexible job shop scheduling problem with transportation robots in Table 1 according to different criteria. 1. 2. 3. 4. 5. 6.

Used approach The number of transport robots (SR to single/MR to transport robots) The flexibility of job shop environment (Yes or No) Optimization criterion The authors Method type (E: Exact method; A: Approximate method; D: Distributed method).

The classical and flexible job shop scheduling problems with transportation time are NP-Hard problems, which show that most of the proposed methods are of the metaheuristics type. Few works used exact methods because they are not effective in finding results within a reasonable time, see Fig. 1. The metaheuristics have shown their efficiency as always for combinatorial problems as mentioned in [32– 38] and the most used methods are based on genetic algorithm or tabu search. The multi-agent models are also used to solve the JSPT and are also based on genetic

Fig. 1 Percentage of each category of approaches studied

Fig. 2 Mono-objective versus multi-objective

Distributed, Classical and Flexible Job Shop Scheduling Problem . . .

135

Fig. 3 Hybrid versus non-hybrid

algorithms. We can note that the most used works have not studied the flexibility of jobs because the difficulties of this constraint. We can also note that the most used methods a mono objective as an optimization criterion, see Fig. 2. Also, the most studied approach used is not hybrid whatever the categorie of used method, see Fig. 3.

4 Conclusion and Perspectives In this paper, we present the different works proposed for the distributed job shop scheduling problem where we present a classification of the works according to four criteria which: The used approach, the number of transport robots (SR to single/MR to transport robots), the flexibility of job shop environment (Yes or No), the optimization criterion and the Method type. As perspectives, we propose to treat the distributed real-time job scheduling problem with transportation robots to schedule, the operations and take into consideration the new operations that arrive to execute and the cases where the machines break down and transport them from one resource to another and from stock radius to machine and finally from machine to an deposit and we need to move the operation many robots and Automated Guided Vehicles (AGV).

136

B. Marzouki et al.

Table 1 A classification of the studied approaches Approach

SR/MR

Flexibility

Optimization criterion

References

Method type

Linear programming

MR

No

Makespan

[4]

E

Branch and bound

MR

No

Makespan

[5]

E

Conic scalarization method

SR

Yes

Makespan

[6]

E

Constraint programming

MR

No

Makespan

[7]

E

A forward propagation heuristic

MR

No

Makespan, Work in process and Inventory holding costs

[8]

A

Tabu search

SR

No

Makespan

[9, 10]

A

Genetic algorithm with a heuristic

MR

No

Makespan

[11]

A

Local search

MR

No

Makespan

[15]

A

Simulated annealing + local search

MR

Yes

Makespan and Buffer management

[12]

A

Genetic algorithm and tabu search

SR

Yes

Makespan and Buffer management

[13]

A

Ant colony optimization algorithm

MR

No

Makespan, Buffer management, VPM,VCM and PC

[16]

A

Time-Lag Heuristic

MR

No

Makespan and Quality of Service

[14]

A

Hybrid imperialist competitive

SR

Yes

Makespan

[17]

A

Iterated greedy insertion

SR

Yes

Makespan

[18]

A

Simulation approach

MR

Yes

Makespan, robot travel distance, time difference with due date and critical waiting time

[19]

A

Chemical reaction optimization

SR

Yes

Makespan

[20]

A

Differential evolution algorithm

SR

Yes

Makespan

[21]

A

Estimation of distribution algorithm and VNS

SR

Yes

Makespan and total energy consumption

[22]

A

PSO+SA

SR

No

makespan

[23]

A

Reinforcement Learning

SR

Yes

Makespan and energy

[24]

A

Mathamatical model

SR

Yes

Makespan

[25]

A

Genetic Algorithm

SR

Yes

Makespan

[26]

A

Multi agent model

MR

No

Makespan

[27]

D

Multi agent model

MR

No

Makespan

[28]

D

Multi agent model+GA+TS

MR

No

Makespan

[29]

D

Multi agent model+GA+TS

MR

Yes

Makespan

[30]

D

Multi agent model+GA+TS

SR

Yes

Makespan

[31]

D

Distributed, Classical and Flexible Job Shop Scheduling Problem . . .

137

References 1. Cohen, M.A., Lee, H.L.: Strategic analysis of integrated production distribution systems: models and methods. Oper. Res. 36(2), 216–28 (1998) 2. Barroso, A.M., Leite, J.C.B., Loques, O.G.: Treating uncertainty in distributed scheduling. J. Syst. Softw. 63, 129–36 (2002) 3. Lenstra, J.K., Rinnooy Kan, A.H.G.: Complexity of vehicle routing and scheduling problems. Networks 11 2, 221-227 (1981) 4. Raman, N., Talbot, F.B., Rachamadgu, R.V.: Simultaneous scheduling of machines and material handling devices in automated manufacturing. In: Proceedings of the 2nd ORSA/TIMS Conference on Flexible Manufacturing Systems, pp. 455–466 (1986) 5. Pundit, R., Palekar, U.: Job shop scheduling with explicit material handling considerations. University of Illinois at Urbana-Champaign, Dept. of M. and I.E, Technical report (1990) 6. Deliktas, D., Orhan, T., Ozden, U.: A flexible job shop cell scheduling with sequence-dependent family setup times and intercellular transportation times using conic scalarization method. Int. Trans. Oper. Res. 26, 2410–2431 (2019) 7. Ham, A.: Transfer-robot task scheduling in job shop. Int. J. Prod. Res. (2020). https://doi.org/ 10.1080/00207543.2019.1709671 8. Anwar, M.F., Nagi, R.: Integrated scheduling of material handling and manufacturing activities for just-in-time production of complex assemblies. Int. J. Prod. Res. 36, 653–681 (1998) 9. Hurink, J., Knust, S.: A tabu search algorithm for scheduling a single robot in a job-shop environment. Discrete Appl. Math. 119, 181–203 (2002) 10. Hurink, J., Knust, S.: Tabu search algorithms for job-shop problems with a single transport robot. Eur. J. Oper. Res. 162, 99–111 (2005) 11. Tamer, A.F., Nassef, A.O., Kamal, B.A., Hassan, M.F.: A hybrid GA/heuristic approach to the simultaneous scheduling of machines and automated guided vehicles. Int. J. Prod. Res. 42, 267–281 (2004) 12. Deroussi, L., Norre, S.: Simultaneous scheduling of machines and vehicles for the flexible job shop problem. In: International Conference on Metaheuristics and Nature Inspired Computing, pp. 1–2 (2010) 13. Zhang, Q., Manier, H., Manier, M.A.: A genetic algorithm with tabu search procedure for flexible job shop scheduling with transportation constraints and bounded processing times. Comput. Oper. Res. 39, 1713–1723 (2012) 14. Gondran, M., Huguet, M.J., Lacomme, P.H., Quilliot, A., Tchernev, N.: A Dial-a-Ride evaluation for solving the job-shop with routing considerations. Eng. Appl. Artif. Intell. 74, 70–89 (2018) 15. Lacomme, P., Larabi, M., Tchernev, N.: A disjunctive graph for the job-shop with several robots. In: The 3rd Multidisciplinary International Conference on Scheduling: Theory and Applications, pp. 285–292 (2007) 16. El Khoukhi, F., Lamoudan, T., Boukachour, J., Alaoui, A.E.H.: Ant colony algorithm for justin-time job scheduling with transportation times and multirobots. ISRN Appl. Math. 1-19 (2011) 17. Karimi, S., Ardalan, Z., Naderi, B.: Scheduling flexible job-shops with transportation times: Mathematical models and a hybrid imperialist competitive algorithm. Appl. Math. Model 41, 667–682 (2017) 18. Bekkar, A., Ghalem, B., Bouziane, B.: Iterated greedy insertion approaches for the flexible job shop scheduling problem with transportation times constraint. Int. J. Manuf. Res. 14, 1 (2019) 19. Xu, Y., Sahnoun, M., Ben Abdelaziz, F.: A simulated multi-objective model for flexible job shop transportation scheduling. Ann. Oper. Res. https://doi.org/10.1007/s10479-020-036000 (2020) 20. Marzouki, B., Belkahla Driss, O., Ghedira, K.: Improved chemical reaction optimization for distributed flexible job shop problem with transportation times. IFAC-PapersOnLine 55(10), 1249–1254 (2022)

138

B. Marzouki et al.

21. Wu, X., Liu, X., Zhao, N.: An improved differential evolution algorithm for solving a distributed assembly flexible job shop scheduling problem. Memetic Comp. 11, 335–355 (2019) 22. Du, Y., Li, J., Luo, C., Meng, L.: A hybrid estimation of distribution algorithm for distributed flexible job shop scheduling with crane transportations. Swarm Evol. Comput. 62, 100861 (2021) 23. Fontes, D.B.M.M., Homayouni, S.M., Gonc˛alves, J.F.: A hybrid particle swarm optimization and simulated annealing algorithm for the job shop scheduling problem with transport resources. Eur. J. Oper. Res. (2022) 24. Du, Y., Li, J., Li, C., Duan, P.: A reinforcement learning approach for flexible job shop scheduling problem with crane transportation and setup times. IEEE Trans. Neural Netw. Learn. Syst. (2022). https://doi.org/10.1109/TNNLS.2022.3208942 25. Jiang, T., Zhu, H., Liu, L., Gong, Q.: Energy-conscious flexible job shop scheduling problem considering transportation time and deterioration effect simultaneously. Sustain. Comput. Inform. Syst. 35 (2022) 26. Imran Ali, C., Amer Farhan, R., Isam A-Q, E., Mohamed, A., Muhammed, U., Mohamed, B., Attia, B.: Integrated scheduling of machines and automated guided vehicles (AGVs) in flexible job shop environment using genetic algorithms. Int. J. Ind. Eng. Comput. 13, 343–362 (2022) 27. Erol, R., Sahin, C., Baykasoglu, A., Kaplanoglu, V.: A multi-agent based approach to dynamic scheduling of machines and automated guided vehicles in manufacturing systems. Appl. Soft Comput. 12, 1720–1732 (2012) 28. Petrovi’c, M., Miljkovi’c, Z., Babi’c, B.: Integration of process planning, scheduling, and mobile robot navigation based on triz and multi-agent methodology. FME Trans. 41(2), 120– 129 (2013) 29. Nouri, H.E., Belkahla Driss, O., Ghedira, K.: Hybrid metaheuristics for scheduling of machines and transport robots in job shop environment. Appl. Intell. 45, 808–828 (2016) 30. Nouri, H.E., Belkahla Driss, O., Ghedira, K.: Simultaneous scheduling of machines and transport robots in flexible job shop environment using hybrid metaheuristics based on clustered holonic multiagent model. Comput. Ind. Eng. 102, 488–501 (2016) 31. Nouri, H.E., Belkahla Driss, O., Ghedira, K.: Controlling a single transport robot in a flexible job shop environment by hybrid metaheuristics. LNCS Trans. Comput. Collective Intell. 28(1), 93–115 (2018) 32. Chaouch, I., Driss, O.B., Ghedira, K.: A survey of optimization techniques for distributed job shop scheduling problems in multi-factories. In: Silhavy, R., Senkerik, R., Kominkova Oplatkova, Z., Prokopova, Z., Silhavy, P. (eds.) Cybernetics and Mathematics Applications in Intelligent Systems. CSOC 2017. Advances in Intelligent Systems and Computing, vol. 574. Springer, Cham (2017) 33. Chaouch, I., Driss, O.B., Ghedira, K.: A review of job shop scheduling problems in multifactories. Int. J. Oper. Res. 38(2), 147–165 (2020) 34. Marzouki, B., Belkahla Driss, O.: Multi agent model based on chemical reaction optimization for flexible job shop problem. In: International Conference on Computational Collective Intelligence, vol. 9329, pp. 29–38 (2015) 35. Marzouki, B., Belkahla Driss, O., Ghedira, K.: Multi agent model based on chemical reaction optimization with greedy algorithm for flexible job shop scheduling problem. Procedia Comput. Sci. 112, 81–90 (2017) 36. Marzouki, B., Belkahla Driss, O., Ghedira, K.: Multi-agent model based on combination of chemical reaction optimisation metaheuristic with Tabu search for flexible job shop scheduling problem. Int. J. Intell. Eng. Inform. 6(3/4), 242–265 (2018) 37. Marzouki, B., Belkahla Driss, O., Ghedira, K.: Solving distributed and flexible job shop scheduling problem using a chemical reaction optimization metaheuristic. Procedia Comput. Sci. 126, 1424–1433 (2018) 38. Marzouki, B., Belkahla Driss, O., Ghedira, K.: Decentralized Tabu searches in multi agent system for distributed and flexible job shop scheduling problem. In: IEEE/ACS 14th International Conference on Computer Systems and Applications (AICCSA), pp. 1019–1026 (2017b)

Business Economics

Proposal of Bicycle Sharing Operation System by Multi-agent Reinforcement Learning Using Transfer Learning Kohei Yashima and Setsuya Kurahashi

Abstract In this research, we propose a new autonomous bicycle sharing management system by local residents using the MARL (multi-agent reinforcement learning) model that adopts DQN (Deep Q-Network) with four stations as one group. In addition, we will define similar environments by assigning demand-based labels to stations in order to adapt to changes in the environment, such as the addition of more stations, and to confirm the effectiveness of efficient transfer learning. As a result of the experiment, the proposed model allowed multiple reinforcement learning agents to learn cooperative behavior and avoid a situation in which the number of remaining bicycles reaches zero. Furthermore, the performance of the model with and without transfer learning was compared, and the learning speed was higher when transfer learning was used, indicating the effectiveness of the model and the possibility of efficient service operation.

1 Introduction In recent years, there have been proposals for the development of convenient public transportation and energy-saving measures to reduce CO2 emissions, and as part of these efforts, the introduction of bicycle sharing has been promoted in many areas. As the demand for bicycle sharing changes, bicycles are unevenly distributed at specific stations. One of the major problems in operating a shared-cycle business is the operational cost, such as labor and delivery costs for trucks that make rounds to collect and distribute bicycles so that they are not unevenly distributed among the stations. And user demand changes depending on location (e.g., office areas or tourist spots), time of day, and weather conditions. In addition, it is difficult to conduct vehicle dispatching in response to a continuously changing environment, such as the addition of new stations or the expansion of service areas. K. Yashima (B) · S. Kurahashi University of Tsukuba, 1-1-1 Tennodai, Tsukuba, Ibaraki, Japan e-mail: [email protected] S. Kurahashi e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_13

141

142

K. Yashima and S. Kurahashi

Therefore, this study proposes an operation method to solve the uneven distribution of bicycles by requesting the redistribution of bicycles to local residents. Local residents are defined as residents who use bicycle sharing services in their daily lives, but who are not using the services at the time of the request for redistribution, and who are at home or out for shopping. The request for redistribution to local residents is made using Deep Q-Network (DQN), which is one of the reinforcement learning methods. The DQN agent manages four stations as a group, and aims to eliminate stations where the number of bicycles remaining is zero by performing multi-agent reinforcement learning (MARL) in cooperation with neighboring DQN agents. We also discuss the effectiveness of transferring the learning results of a DQN agent to neighboring DQN agents in order to adapt to changes in the environment, such as the addition of new stations. Compared to the conventional operation method using trucks, this method is expected to eliminate the uneven distribution of stations and reduce CO2 emissions, and is expected to contribute to the promotion of sharing economy and decarbonization, which are closely linked to the SDGs.

2 Related Work 2.1 User-Based Approach of Bicycle Sharing One approach to solve uneven station distribution is to request users to collaborate in changing their lending and returning stations. Singla et al. [1] propose an approach that attempts to solve uneven distribution by offering rewards to users to request their collaboration in changing lending and returning stations. In an experiment with a company that provides a bicycle sharing service, it is noted that the acceptance rate of the offer for all participants was about 60%, and that there was bias in the areas and times of the day when the offer was accepted. Users’ decisions on collaboration requests are complicated, but the incentive price offered and the travel costs related to changing the lending and returning stations are considered to be the major factors.

2.2 Transfer Reinforcement Learning Transfer learning is the reuse of knowledge learned in one domain for learning in a different domain [2]. Transfer learning in reinforcement learning is expected to have the following effects: jump-start, learning speed improvement, and asymptotic improvement [3]. On the other hand, if the relevance between the source and target domains is low, negative transfer may worsen learning performance [4]. Various methods of knowledge reuse among agents using transfer learning have been discussed in MARL, where multiple reinforcement learning agents learn simultaneously [5].

Proposal of Bicycle Sharing Operation System …

143

Xiao et al. [6] propose an approach that combines Q-learning, one of the reinforcement learning [7] methods, and transfer learning to solve uneven station distribution. Q-learning maintains a Q table Q(s, a) that stores the Q value of each action a for each state s. The Q table Q(s, a) is a set of Q values for each state. Each station is a Q-learning agent, which attempts to solve uneven distribution by directing bicycles to move according to the number of remaining stations. The knowledge repository is used for knowledge transfer. After a certain number of episodes, the average value of each agent’s Q table is stored in the knowledge repository. Each agent selects the action with the highest Q value by referring to the knowledge repository and its own Q table, hence realizing knowledge sharing and transfer among agents. Experimental results show that using the knowledge repository for transfers improved the rate of uneven distribution solved, leading to an improvement in performance. On the other hand, the demand for each station is randomly generated using random numbers, and the results were not verified for hourly demand changes such as daytime and nighttime.

3 Proposed Method In this study, we employ a DQN agent with four stations as a group, which is a midpoint between the business operator and local residents. DQN is a model-free reinforcement learning method that does not require knowledge of the environment. It is easy to adjust and can be operated with a certain level of performance at a low cost because it uses actual experience gained from the environment. In addition, the proposed method defines similarity based on the combination of demand trends for one group of four stations and transfers learning results to similar groups in different areas. This is expected to provide high generalization performance and low computational load even when deployed in a wide area where additional stations are expected to be added.

3.1 Simulation Environment As a schematic of the reference environment for the simulation, Fig. 1 shows the location of the station and the DQN agent. Initially, 10 bicycles are assumed to be installed at each station. Users and local residents are assumed to exist in the simulation environment, and requests for dispatching bicycles to local residents are notified by a smartphone application, and shopping coupons that can be used in the local area are provided to local residents as incentives for responding to such requests. The demand for each station is shown in Fig. 2.

144

K. Yashima and S. Kurahashi

Fig. 1 Schematic of the simulation environment

Fig. 2 Number of bicycles remaining at each station per hour

3.2 Cooperative Reinforcement Learning The DQN agent has a group of four station environments and uses the ε-greedy method, which acts randomly with probability ε, to learn the value Q(s, a) for selecting action a in a state s in an exploratory way. The parameter ε decays gradually from an initial value of 1 to a final value of 0.0005. The other parameters, learning rate α is set to 0.001 and discount rate γ is set to 0.95. The fully connected neural network consists of two intermediate layers with 24 nodes, and uses the ReLU (Rectified Linear Unit) as the activation function. The input state s and output action a of each DQN agent are shown below. s1: Number of stations remaining at each station s2: Total incentive for all stations paid to local residents per hour a1: No dispatch request from any station a2: Request for 3 dispatches from station X 1 to X 0 a3: Request for 3 dispatches from station X 2 to X 0

Proposal of Bicycle Sharing Operation System …

145

a4: Request for 3 dispatches from station X 3 to X 0 a5: Request for 3 dispatches from station X 0 to X 1 a6: Request for 3 dispatches from station X 2 to X 1 ... a13: Request for 3 dispatches from station X 2 to X 3 The state s1 is an observable state in order to solve the stations such that the number of stations remaining becomes zero, and the state s2 is an observable state in order to reduce the overall incentive paid by the business operator. The action a selects one action from 13 options, assuming that the four stations in the group managed by each DQN agent are X 0, X 1, X 2, and X 3. As rewards for the DQN agents, we designed the system to give a large negative point when the number of bicycles remaining at any of the stations reaches 0, and a small negative point for each hourly dispatch request.

3.3 Selection of a Transfer Site A label is assigned to the demand trend for each station. Low label is assigned to a target station for which the remaining number of stations is less than or equal to 0 throughout the day’s demand. Other target stations are labeled High. Below 0: Low More than 1: High In selecting the transfers, the similarity of the transfers is defined by the combination of the labels of the four stations, and the transfers are made to combinations of the same labels. Compared to the case where similarity is not defined, negative transfers do not occur and good performance is expected.

4 Experiments In all experiments, one step is defined as the result of all agents selecting one action every hour, and 17 actions can be selected in sequence from 7:00 AM to 23:00 PM, so the maximum number of steps is 17. However, since the number of steps and the environment are reset when the number of remaining cars at any station reaches 0, there are cases where the number of steps does not reach 17 and the environment is reset, meaning that the number of steps of 17 means that the number of remaining cars does not reach 0 until 23:00 PM during all time periods. One episode is defined as when the number of remaining cars at any station reaches 0 or the number of steps reaches 17. When one episode elapses, the number of steps and the environment are reset and a new episode begins.

146

K. Yashima and S. Kurahashi

Fig. 3 The results of Experiment 1 without transfer are shown in (a), and the results with transfer are shown in (b)

4.1 Experiment 1_Transfer Learning with Homogeneous Labels We confirm the effectiveness of transfer learning with homogeneous labels. DQN2 of Group B and DQN1 of Group A have the same label combination. To confirm the behavior of the transfer learning method, which uses labels to select transfer targets, we transfer the results of 3000 episodes of learning by DQN1 to DQN2. We will compare the performance of DQNs with and without transfer to confirm how the performance changes. The results without transfer are shown in Fig. 3a. The results with transfer are shown in Fig. 3b. The first axis of the vertical axis represents the total number of points, which is the reward at the end of the episode, the second axis represents the value of ε in the ε-greedy method, and the horizontal axis represents the number of episodes. Without transfer was observed around 700 episodes, and operations in which the number of remaining stations did not reach 0 were observed. On the other hand, with transfer was observed from around 150 episodes earlier, and the operation in which the number of remaining stations did not become zero was stably learned after 400 episodes, confirming the expected improvement in learning speed as an effect of transfer learning.

4.2 Experiment 2_Transfer Learning with Heterogeneous Labels We confirm the behavior of transfer learning with heterogeneous labels. DQN3 in Group C and DQN1 in Group A are different label combinations. The results of 3000 episodes of learning by DQN1 are transferred to DQN3. We will compare the performance of DQNs with and without transfer to confirm how the performance changes. The results without transfer are shown in Fig. 4a. The results with transfer are shown in Fig. 4b.

Proposal of Bicycle Sharing Operation System …

147

Fig. 4 The results of Experiment 2 without transfer are shown in (a), and the results with transfer are shown in (b) Fig. 5 Overall environment

Without transfer was observed from around 1000 episodes, in which the number of remaining stations did not reach 0. On the other hand, With transfer was observed from around 600 episode earlier, and it was confirmed that even if the labels were heterogeneous, transferring them was more effective in improving the learning rate.

4.3 Experiment 3_Transfer Learning Between Homogeneous and Heterogeneous Labels We confirm the behavior of transitions between homogeneous and heterogeneous labels. To confirm the behavior, we create DQN4 in group D, which is the same label combination as DQN3 in group C. A new schematic of the simulation environment is shown in Fig. 5. DQN2 to DQN4 is heterogeneous transfer and DQN3 to DQN4 is homogeneous transfer, and the results are compared when DQN4 is learned without transfer and when DQN2 and DQN3 are transferred. The results without transfer are shown in Fig. 6. The results with a heterogeneous transfer in Fig. 7a. The results with a homogeneous transfer in Fig. 7b.

148

K. Yashima and S. Kurahashi

Fig. 6 The results of Experiment 3 without transfer is shown

Fig. 7 The results of Experiment 3 with heterogeneous transfer are shown in (a), and the results with homogeneous transfer are shown in (b)

The homogeneous labels learned operations in which the number of remaining bicycles did not reach zero earlier, confirming that the transfer learning to the homogeneous labels was more effective than to the heterogeneous labels.

4.4 Experiment 4_Cooperative Transfer Reinforcement Learning We confirm that the three groups, DQN1, DQN2, and DQN3, can cooperate to prevent the number of remaining bicycles from reaching zero. In addition, the effectiveness of transfer learning is confirmed by comparing the results with and without transfers between homogeneous labels, i.e., DQN1 to DQN2 and DQN4 to DQN3. The results without transfer are shown in Fig. 8a. The results with transfer are shown in Fig. 8b. The experimental results confirmed that the DQNs were able to cooperate with each other and learn operations that avoid the number of remaining stations reaching zero in the latter half of the episode, regardless of whether the transfer learning was performed or not. In addition, without transfer confirmed that the operation in which

Proposal of Bicycle Sharing Operation System …

149

Fig. 8 The results of Experiment 4 without transfer are shown in (a), and the results with transfer are shown in (b)

the number of remaining stations does not reach zero is observed from around 2000 episodes, while with transfer confirmed it even earlier, from around 1700 episodes, confirming that transfer learning is more effective in improving the learning speed.

4.5 Discussion In this study, we considered that negative transfer could be avoided by assigning Low and High labels to stations based on whether the number of remaining stations reaches 0 or not, and using a combination of these labels to determine the transfer target. In Experiment 3, we confirmed that transitions between homogeneous labels were better than those between heterogeneous labels. On the other hand, in Experiment 2, we confirmed that transfers between heterogeneous labels were also effective, indicating that negative transfers did not occur. Although the label combinations were different, the large demand trends were similar, resulting in effective transfers. In addition, the demand for each station is complicated, and due to the difference in the time period when the number of remaining stations is zero, there are cases where the similarity of transfer targets cannot be sufficiently defined by the two types of demand, low and high, and the number of remaining stations is remains zero, and so on. It is necessary to consider how to express the index for selecting the transfer targets as an important issue to be solved in the future. In Experiment 4, each DQN agent learns cooperative behavior and is able to achieve an operation in which the number of remaining bicycles does not reach 0. However, learning is unstable, and the reward design and the input and output of each agent also need to be considered.

5 Conclusion In this study, we confirmed the possibility of a bicycle sharing operation system by local residents using the MARL model, which employs DQNs with four stations as

150

K. Yashima and S. Kurahashi

a group, and which can respond to the expansion of stations. In the case of a small area and demand such as in the simulation environment, it was shown that each DQN agent can cooperate with each other so that the number of remaining bicycles does not reach zero. The use of transfer learning also improved the performance of the DQN agents, and showed the possibility of expanding the number of stations and the area. However, it is expected that as the number of DQN agents increases, the number of combinations of states and actions becomes huge, and the actions involved in cooperation become more complicated. Therefore, a multi-level hierarchical structure is considered to be necessary, such as making it easier to find cooperative actions and establishing an agent that manages each DQN from a higher level. we would like to consider the configuration and experimentation in the future.

References 1. Singla, A., Santoni, M., Bartok, G., Mukerji, P., Meenen, M., Krause, A.: Incentivizing users for balancing bike sharing systems. In: Twenty-Ninth AAAI Conference on Artificial Intelligence, pp. 723–729 (2015) 2. Kamishima, T.: Transfer learning. Jpn. Soc. Artifi. Intell. 25(4), 572–580 (2010) 3. Taylor, M., Stone, P.: Transfer learning for reinforcement learning domains: a survey. J. Mach. Learn. Res. (2009). https://doi.org/10.1633/1685 4. Hitoshi, K.: Reinforcement and Transfer Learning in Python. Morikita Publishing (2022) 5. Silva, D., Leno, F., Costa, A.: A survey on transfer learning for multiagent reinforcement learning systems. J. Artifi. Intell. Res. 64, 645–703 (2019) 6. Xiao, I.: A Distributed Reinforcement Learning Solution With Knowledge Transfer Capability for A Bike Rebalancing Problem, arXiv preprint arXiv:1810.04058 (2018) 7. Sutton, R., Barto, A.: Reinforcement Learning. Morikita Publishing (2000) 8. Mnih, V., Kavukcuoglu, K., Silver, D., Graves, A., Antonoglou, I., Wierstra, D., Riedmiller, M.: Playing Atari with Deep Reinforcement Learning. arXiv preprint arXiv:1312.5602 (2013) 9. Mnih, V., Kavukcuoglu, K., Silver, D., et al: Human-level control through deep reinforcement learning. Nature (London) 518(7540), 529–533 (2015) 10. Dotterl, J., Bruns, R., Dunkel, J., Ossowski, S.: Towards dynamic rebalancing of bike sharing systems: an event-driven agents approach. Prog. Artifi. Intell.: EPIA 2017, 309–320 (2017) 11. Nguyen, T., Nguyen, N., Nahavandi, S.: Deep reinforcement learning for multiagent systems: a review of challenges, solutions, and applications. IEEE Trans. Cyber. 50(9), 3826–3839 (2020) 12. Hafiz, A., Bhat, G.: Deep Q-Network Based Multi-agent Reinforcement Learning with Binary Action Agents. arXiv preprint arXiv:2008.04109 (2020)

Change in Centrality and Team Performance: Inverse Relation Between Manager and Non-manager Communication Hitomi Inagaki and Setsuya Kurahashi

Abstract The purpose of this study is to examine the impact of changes in centrality on performance and to gain a better understanding of performance management within organizations. The communication records of 453 employees from a manufacturing company were analyzed, and 36 teams’ performance was used to compute network indicators using social network analysis. The findings indicate that a manager’s betweenness centrality has an impact on team performance. The results also revealed an inverse correlation between the manager’s betweenness centrality and the non-manager’s degree ratio in meeting communication in high-performing teams.

1 Introduction In recent years, the interdependence of the global economy, geopolitical risks, and environmental concerns have become complexly intertwined, resulting in an increase in uncertainty in the business environment. In such a rapidly changing and uncertain environment, companies and their employees must leverage both internal and external networks to access new information, knowledge, and form diverse relationships to achieve success. The study of employees’ social networks and their performance in organizations has been carried out for a long time [1, 2]. In this setting, activating employees’ changing networks has attracted considerable interest for organizational management [3, 4]. Thus, this study explores how the change in centrality of employees who have connections inside and outside the company over a period of one year affect team performance. We use communication log to examine how manager and non-managers’ communication is related and how it affects performance. This paper is organized into six sections. Section 2 provides an overview of related research on network centrality and performance in companies. Section 3 describes H. Inagaki (B) · S. Kurahashi University of Tsukuba, 1-1-1 Tennodai, Tsukuba, Ibaraki, Japan e-mail: [email protected] S. Kurahashi e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_14

151

152

H. Inagaki and S. Kurahashi

the research methodology and variables used in this study, as well as the calculation method applied to the network metrics. Section 4 shows the results of the regression analysis between the network centrality and performance. Section 5 discusses the implications of the results. Section 6 discusses the conclusion and limitations of the study.

2 Previous Research 2.1 Social Network Analysis Social network analysis is an approach that measures interactions and the social structures using graph theory. Relationships typically comprise nodes and edges that represent the connections between nodes. When V and E respectively refer to a set of nodes and edges, such network G is defined as G = (V, E) [5]. We examine the relationship between centrality and performance using social network analysis.

2.2 Performance and Centrality in Companies There have been many studies regarding the relationship between centrality of leaders/managers and performance (Ref. [6–8] etc.). For example, degree centrality, which counts the number of paths that emanate from an organizational actor [9], is generally considered to indicate a leader’s popularity. It is said to have a positive effect on team leader’s degree centrality and team performance [6]. A study of 19 teams in a manufacturing organization showed that teams with leaders whom many subordinates sought advice from had lower conflict within the team and high viability [10]. On the other hand, some studies have pointed out the disadvantages of high degree centrality. As the direct linkages increase, maintaining them may consume more personal resources [11]. Furthermore, having strong connections with others may lead to individual behavior being restricted by roles defined by those connections [12]. Additionally, betweenness centrality, which represents the degree of mediating between individuals, is a strong predictor of leadership and being perceived as a leader. Leaders with high betweenness centrality serve as intermediaries between different teams and coordinate work and information flow within organizations [6]. On the other hand, a high level of betweenness centrality may also result in more conflicting demands from teams and individuals not connected to them [6]. The eigenvector centrality evaluates nodes with more connections [5]. It is said to help avoid the cost associated with high degree centrality and betweenness centrality positions [6]. A study using network data from a financial services firm’s sales

Change in Centrality and Team Performance …

153

division found a positive relation between the height of eigenvector centrality in the relationship between team leaders and colleagues and their reputation for leadership from subordinates [13]. As highlighted above, past research has established a correlation between the centrality of leaders and managers and their performance. Nevertheless, previous company-focused studies commonly rely on data obtained from a single point in time or aggregated data. This study aims to address the limitations of previous research by examining the dynamic change in employee centrality over time and its relationship with performance using communication log data.

3 Methodology 3.1 Data Outline The data used in this study consists of 453 employees of a major Japanese manufacturer. There were a total of 36 teams. Communication log data of meetings (both face-to-face and online) were collected over a period of 12 months, from April 2021 to March 2022 (the company’s fiscal year) for these employees. The data only captures who communicated with whom and does not include the content of conversations.

3.2 Variables 3.2.1

Independent Variable

We utilized the end-of-term evaluations of the 453 employees as a measure of team performance. The evaluations were divided into four categories: S, A, B, and C, with S being the highest and C being the lowest. To obtain numerical values, S was assigned 4, A was 3, B was 2, and C was 1. We then calculated the average of each employee’s evaluations over the past three years and used the average of these values for each team to determine the team’s performance.

3.2.2

Dependent Variables

We removed variables with high Variance Inflation Factor (VIF) to eliminate the impact of multicollinearity [14, 15]. As a result, degree centrality was excluded and eigenvector centrality and betweenness centrality were used as independent variables. Betweenness centrality represents how many people are mediated in the network, meaning the degree to which a node is located on the shortest path between other nodes [5].

154

H. Inagaki and S. Kurahashi

Cb (i) =

 g jk (i) g jk i= j=k

(1)

where g jk (i) is the number of shortest paths between node j and node k that pass through node i, and g jk is the number of shortest paths between node j and node k. The eigenvector centrality measures the influence of a node in a network, as represented by the centrality index of the first eigenvector of the adjacent matrix [5]. The score of eigenvector centrality for vertex i is defined as follows: [16]. xi =

1 ai j x j λ j∈V

(2)

Here, λ is the eigenvalue of a positive definite matrix A. The average values of these two centralities were calculated for each individual for one month, and then the median and standard deviation of the values were calculated over a 12-month period.

3.2.3

Control Variables

Although numerous variables could affect performance, not all could be included. This study used density, efficiency, and employee attributes as control variables. Network centrality depends on the density and efficiency of each team, so we used density as a control variable [17]. Density is the ratio of the actual number of edges to the number of all edges [5]. The formula can be expressed as 2m/{n(n − 1)} where the maximum number of possible edges in an undirected graph with n is n(n − 1)/2 and the number of edges in the graph is m. Burt proposed network efficiency as a concept to describe a lack of structural holes [2]. Efficiency refers to the percentage of nodes that do not overlap with each other; The higher this value, the less overlap between nodes; meaning the more diverse the information sources. Network efficiency requires both redundancy and effective size [18, 19]. When we denote t as the number of connections other than those connected to a node (we call it the ego), redundancy can be expressed as redundancy = 2t/n. The effective size is the total number of nodes other than the ego minus duplicates, so it is defined as effective size Si = n − (2t/n). If we denote the ego’s degree as ki , then efficiency is the effective size divided by the ego’s degree. Finally, the attributes used in this study include average age of the team, gender dummy, education level (high school, bachelor’s, graduate and above), percentage of managers, proportion of hire types (mid-career or fresh graduate), job type dummy (business planning and sales), and team size.

Change in Centrality and Team Performance … Table 1 The average and standard deviation of each variable Variables Mean SD Variables Dependent variable Controls Evaluation 2.22 0.28 Age Network variables Male (dummy) Eigenvector centrality 0.04 0.01 Academic record Betweenness centrality 0.00 0.00 Manager (dummy) Network controls Hire type (dummy) Density 0.98 0.07 Jobtype: planning (dummy) Efficiency 0.42 0.04 Jobtype: sales (dummy) Team size

155

Mean

SD

44.99 0.77 2.39 0.12 0.84 0.13

4.80 0.23 1.26 0.14 0.19 0.31

0.03

0.17

11.25

9.49

4 Results 4.1 Centrality and Performance First, we show the mean and standard deviation of team performance, network indicators (centrality), and control variables in Table 1. We then performed a multiple regression analysis using the ordinary least squares (OLS) method, with team performance as the dependent variable. To explore the relationship between centrality and team performance, we calculated three models. Model 1 used the centrality values calculated at one point in time, specifically the final month of the year, as the independent variable. Model 2 used the median centrality values calculated over the course of the year as the independent variable. Lastly, Model 3 included both the median and standard deviation of centrality over the course of a year as independent variables. We used these three models to determine the extent to which centrality was associated with team performance, while controlling for other variables. The results are shown in Table 2. The best-fitting model was the one that included the standard deviation of centrality over 12 months (Model 3). The results showed that eigenvector centrality had the highest positive impact on performance among the independent variables. The standard deviation of betweenness centrality was also found to have a positive impact on team performance in Model 3.

156

H. Inagaki and S. Kurahashi

Table 2 The regression results Variables Constant Network variables Eigenvector centrality (median) Between centrality (median) Eigenvector centrality (SD) Between centrality (SD) Network control Density Global efficiency Control Age Male (dummy) Academic record Manager (dummy) Hire type (dummy) Jobtype: planning (dummy) Jobtype: sales (dummy) Team size Adj. R-squared BIC

Model 1

Model 2

Model 3

-1E-15

-3E-16

-3E-16

0.97** −0.12

0.72* 0.12

0.93** −0.12 0.41 0.38*

−0.69** −0.39†

−0.37† 0.09

−0.58* −0.01

−0.32 0.68* −0.78* −0.44* 0.32† 0.09 0.18 −0.25 0.32 122

−0.35 0.52† −0.73† −0.35† 0.27 −0.10 0.15 0.01 0.23 124

−0.58* 0.46† −0.56 −0.23 0.47* −0.33 0.04 −0.04 0.39 110

† p 10%, subjective judgment must be revised [12].

An Analytic Hierarchy Process for Student Motivation

283

Fig. 1 Cognitive Map based on the authors’ personal experiences

3 Cognitive Map The field of cognitive map design research, as first introduced in Robinson’s “The Look of Maps,"[10] has significantly impacted the understanding of human brain manipulation in the 20th century. In this study, a cognitive map was designed based on the knowledge gained through lecture and coursework in Pedagogy and Psychology in Higher Education, as depicted in Fig. 1. The various steps involved in the design process are outlined in Table 1. This study also employs aspects of the Analytic Hierarchy Process (AHP), which will be described in Sect. 2.1. Table 1 highlights the role of each method depicted in Fig. 1, further explaining the design process of the cognitive map. • • • •

Intermediate : In between two methods Loop : In between two methods and also goes to main modify whole map Criteria : Decision makers depended on the problem type (AHP [6]) Alternative : Availability option (AHP [6]).

Our primary objective is to enhance productivity in the classroom. To achieve this, we employed various methods, which are depicted in Fig. 1. This included manipulation of data such as student profiles and student feedback to inform our decision-making process using the Analytical Hierarchy Process (AHP) decision matrix.

3.1 Relation Between Mind Map and AHP Here we simplified my map (Fig. 1) to AHP style where we define criteria and alternative very simple style shown in Fig. 2.

284

R. Islam et al.

Table 1 Explanation of Cognitive Map with AHP feature Name/ Methods Role/ AHP feature Comment Visual Presentation

Criteria Intermediate

Tools Using note card Image Slide Auditory Audio Video Meme

Intermediate Alternative Alternative Alternative Criteria Alternative Alternative

Student collaboration software Tactile / Kinesthetic Hand or body movement Break Study group Student profile Communication Verbal Discussion Direct experience Refractive learning Student feedback Classroom activity

Alternative Criteria Alternative Alternative Alternative Loop Intermediate Criteria Intermediate Intermediate Intermediate Loop Intermediate

Explain in 1.2 It helped to compress learning malarial also easy to presentable – Making the cards makes active learner Showing image is helpful for visual learning Combination of picture and text Explain in 1.3 Helps better to understand slide Form of online content, such as an image, video, or text, that is widely and quickly shared on the internet with minor modifications, often for comedic purposes Collaboration tools using internet Explain in 1.4 Helps to motivate blood flow Helps to refresh mind and body discuss with idea helps to spread faster Explain in 1.8 Help to communicate each other to grow Explain in 1.5 Help to communicate each other to grow Explain in 1.6 Explain in 1.7 Explain in 1.9 Any type of class work like quiz, project , exam or discussion

Simplified criteria and alternative are define bellow. • Criteria – – – –

Visual Auditory Tactile Verbal

• Alternative – Image – Slide

An Analytic Hierarchy Process for Student Motivation

285

Fig. 2 Simplified mind map for analysis

– – – – – – –

Audio & Video Meme Student Collaboration Software Using note card Hand or body movement Break Study group.

4 Case Study This sample case study bases on student feedback and student profile. Assume a decision matrix bases on student’s feet back and we analyze bases on that matrix. Here CR is very important. If and only if CR is less than 10% then we accept the result. So it helps to minimize the risk of student priority.

4.1 Background of the Problem We analyze AHP methods bases on Sect. 3.1. In this case our bases on student feedback and student profile. That why in every case it may vary. We consider 4 criteria and 9 alternatives here. Here we use AHP online software1 to calculate AHP value.

1

https://bpmsg.com/ahp/ahp-calc.php.

286

R. Islam et al.

Fig. 3 Criteria matrix

4.1.1

Criteria and Alternative Figure

For overview of student in class first criteria matrix (Fig. 3) is formed. Here we use AHP rule . In decision matrix (1,4) = 7 means visual is more importance (7x) than verbal.So in alternative (4, 1) = 1/7 (AHP rule). For select alternative additionally consider criteria matrix and set value to decision matrix and find result which is shown in Fig. 4a–d. Same methods apply for alternatives. Also here additional consider criteria which is shown in alternatives figure.

4.2 Result and Discussion We can use any combination of methods that meets a 50%+ threshold. Efficiency may dictate the amount of methods used.The specific combination of methods used may depend on various efficiency considerations. Using the Song formula [14], we analyzed the results from the visual, auditory, tactile, and verbal alternatives (as depicted in Figs. 4a–d) and found that audio and video had high values of 0.2 while hand or body movements had low values of 0.05. Rest of the alternative with values as flows 0.196259 (Audio and Video) > 0.162536 (Slide) > 0.162342 (Image) > 0.124168 (Meme) > 0.096336 (Student Collaboration Software) > 0.089636 (Using note card ) > 0.059551 (Break) > 0.059153 (Study group) > 0.049146 (Hand or body movement).

An Analytic Hierarchy Process for Student Motivation

287

Fig. 4 Verbal alternative matrix based on Fig. 3 and student feedback and profile

5 Conclusion Finally, it can be concluded that the Analytic Hierarchy Process (AHP) provides a useful method for balancing the presence of high-achieving and under performing students in a university class. Based on the principles of Boyer’s scholarship [4] and other factors, the AHP can be helpful for both teachers and students. The study raises two questions: (1) Is this method helpful for both teacher and student? and (2) How many methods can be introduced in a single class (at the second layer)? The answers to these questions are positive. If the combination of methods selected using the AHP reaches a threshold of 50% or above, it can be considered suitable for a class. The criteria matrix is dynamic and changes based on student profiles and feedback. Both visual and auditory methods, as well as other combinations, have been applied pre and post-COVID and the results suggest that they are most efficient when used within the classroom setting.

288

R. Islam et al.

Limitation Online ed. (COVID-19 era) can be beneficial, but there are limitations. Selfmotivation and organization is key. Interaction with teachers and other students is limited, and tech issues may arise. Written comm. is a crucial aspect and can be difficult for some. Furthermore, our focus is on providing an overview based on the changing values. It is possible that different weights may be assigned to the alternatives.

References 1. Anderson, L.W., Bloom, B.S.: A taxonomy for learning, teaching, and assessing: a revision of Bloom’s taxonomy of educational objectives. Longman (2001) 2. Anderson, L.W., Sosniak, L.A.: Bloom’s Taxonomy. University Chicago Press Chicago, IL (1994) 3. Ausubel, D.P.: The facilitation of meaningful verbal learning in the classroom. Educ. Psychol. 12(2), 162–178 (1977) 4. Boyer, E.L.: From scholarship reconsidered to scholarship assessed. Quest 48(2), 129–139 (1996) 5. Clarke, I., III., Flaherty, T.B., Yankey, M.: Teaching the visual learner: the use of visual summaries in marketing education. J. Mark. Educ. 28(3), 218–226 (2006) 6. de FSM Russo, R., Camanho, R.: Criteria in ahp: a systematic review of literature. Procedia Comput. Sci. 55, 1123–1132 (2015) 7. Johnny C GO, S., Atienza, R.J.: Learning by Refraction: A Practitioner’s Guide to 21st Century Ignatian Pedagogy. Ateneo de Manila University Press (2019) 8. Lee, J., Price, N.: A national sports institute as a learning culture. Phys. Educ. Sport Pedagogy 21(1), 10–23 (2016) 9. LeSuer, R.J.: Incorporating tactile learning into periodic trend analysis using three-dimensional printing. J. Chem. Educ. 96(2), 285–290 (2018) 10. Robinson, A.H.: The look of maps: an examination of cartographic design. Am. Cartographer 13(3), 280–280 (1986) 11. Saaty, T.L.: Decision making with the analytic hierarchy process. Int. J. Serv. Sci. 1(1), 83–98 (2008) 12. Shufeng, J.: The algorithm of mean random consistency index in ahp and its implementation. J. Taiyuan Normal Univ. (Nat. Sci. Ed.) 4 (2006) 13. Sogunro, O.A.: Efficacy of role-playing pedagogy in training leaders: some reflections. J. Manage. Dev. (2004) 14. Song, B., Kang, S.: A method of assigning weights using a ranking and nonhierarchy comparison. Adv. Decis. Sci. (2016) 15. Triantaphyllou, E.: Multi-criteria decision making methods. In: Multi-criteria decision making methods: a comparative study, pp. 5–21. Springer, Berlin (2000)

Training Students as Agile Developers: Team and Role Building Games Paolo Ciancarini and Marcello Missiroli

Abstract Computational Thinking is a skill related to problem solving: it is the competence necessary for applying, assessing, producing an algorithmic solution, and implementing it. Agile values and principles are both an ethic framework and a practical reference for teams of contemporary software developers. We study the combination of Computational Thinking competence with Agile values and principles: we have called the result of this combination Cooperative Thinking. Our hypothesis is that the practice of Cooperative Thinking is especially important for students and practitioners who will work in software development teams. Individual productivity in agile software projects is less important than team productivity, which in turn is influenced by team dynamics. In this paper we describe an approach to practicing Cooperative Thinking that we are experimenting, consisting in coaching groups of developers using team-building games, some well known, some invented ad hoc. We discuss the topic of team building by serious games; we show how to evaluate the performance of a team engaged in playing games to train themselves as agile software developers. We will distinguish games playable online from other games, as during the pandemic we were compelled to organize team building tasks online only. We have developed a GQM schema to evaluate the teamwork during a game involving cooperative thinking.

1 Introduction Computational Thinking is the skill related to problem solving and applying, assessing, producing, and implement an algorithmic solution [6]. It is a fundamental skill that lies at the core of problem-solving, as it is necessary to understand what the problem is before developing a solution according to a specific point of view exploiting any computing agent [10]. We are studying how Computational Thinking and Agile values and principles can be combined in a concept called Cooperative Thinking, defined as the ability to P. Ciancarini (B) · M. Missiroli University of Bologna, Bologna, Italy e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_26

289

290

P. Ciancarini and M. Missiroli

Table 1 List of items relevant for cooperative thinking, from [3] Item Description Complex negotiation Continuous learning Group awareness Group organization Social sensitivity

During design, I like to discuss with people who have different ideas, in order to develop the best solution Programming in team taught me something I didn’t know I like to be part of a software developing team When I work in team, results are better than when I work alone During development, I work fine even with teammates with whom I have personal difficulties

describe, recognize, decompose problems and computationally solve them in teams in a socially sustainable way [3]. We analyzed Cooperative Thinking as a social construct using the technique of Partial Least Square Modeling [16]. Using this technique, we were able to single out some pillars of Cooperative Thinking, which are summarized in Table 1. One of the consequences of focusing our attention to Cooperative Thinking is that team building—namely creating groups of people collaborating in a project with good performance and quality of results—is important. Team building is a process by which a group becomes a team [19]. An expert software developer is not necessarily able to cooperate with other developers to solve complex problems. Moreover, Agile principles are not clearly connected to Computational Thinking, and team-based problem solving is still an open problem. Our research group worked on developing Cooperative Thinking with teams of developers, in high schools, in university courses, and in some professional workshops. We started with minimal teams composed of pairs, then we made experiments with larger groups of people [12]. When people team to solve complex problems, as in software development, social sustainability is essential. In this work we define social sustainability as the understanding of the impacts of process interactions on people engaged in software development. When developers communicate they exploit interaction mechanisms and communication channels whose organization can impact the structure of the code they build: Conway was first to observe this phenomenon that now is defined “the Conway’s law” [5]. Communication channels and eventual impediments to human interaction (e.g. like those we are all experiencing during the lockdown) impact the way software systems are designed or composed. Conway’s observation suggests that it is important to manage the social structure and the communication channels of both the developers and the stakeholders interested in their products. Problem solving based on cooperation should be socially sustainable since a developer should be able not only to perform her own task (e.g., developing some piece of code), she should also be able to effectively interact with her social context (e.g., internal and external project’s stakeholders, laws and regulations, safety requirements, etc.). Training students to be aware and able to manage their social context when involved in problem solving for building software requires the ability to recognize that

Training Students as Agile Developers: Team and Role …

291

team-based software development (programming-in-the-many) is very different from single-developer coding (programming-in-the-small). Thus, social sustainability of software development is an issue complementary to both Computational Thinking and agile principles and values. In this paper we describe an approach to Cooperative Thinking that we are experimenting, consisting in coaching students or even senior developers by using games, some well known, some invented ad hoc, some adapted from commercial games. We are interested in the following research questions: RQ1: Which games can be used to coach teams of software developers to the practice of Cooperative Thinking? RQ2: Which games can be used to practice team building online? RQ3: How can we measure the progresses of a group of people engaged in a team building ludic activity involving Cooperative Thinking? The methodology we use consists in experimenting the games with our students (the authors teach a course with about 90 students each at two different third year Computer Science degree; moreover, the second author teaches in two classes of about 25 students at high school level; we both also sometimes teach in workshops for agile professionals). We usually play some games in class under our direct monitoring, while in some cases we ask the students to play in teams without our control but reporting some data to us. There are several games used by agile coaches, but we have found no systematic description or classification of these activities. More important, we have not found any analysis connected to the topic of team building by gamification and the related entertainment value. Team building is a process evolving across different phases [9, 19], each requiring different activities. Hence, we study in particular how the practice of some Agile-oriented games can been used as a means to teach groups high-school or university students to use some specific agile development methods like XP or Scrum, and especially to train and improve their collective problem solving skills. The structure of this paper is the following: Sect. 2 presents some related works; Sect. 3 discusses the issues typical of team building, meaning how to transform a group of developers in a high performance developing team; Sect. 4 describes some games we have used in our practice; Sect. 5 presents our approach to evaluate the results and their validity. Section 6 presents our conclusions and some future works focusing on the impact of the pandemic on our approach.

2 Related Works Software engineering is a difficult and sometimes boring discipline for some students, so it makes sense to apply some team-oriented gamification to engage students and

292

P. Ciancarini and M. Missiroli

keep alive their attention. Starting from the observation that most software development activities can be “played” by people working as teams, games for Agile training and coaching are becoming very popular to engage groups of software developers [14]. A famous classification scheme for games was introduced by the social scientist Roger Caillois [2]: it includes four dimensions: agon (competition), alea (luck), mimicry (imitation), ilynx (vertigo). Combinations are also valid: roll-play combines agon and alea, role-play combines agon and mimicry, and so on. In games for software developers we exploit especially mimicry as simulation, including role-play, that is combined with some agon as competition, see for instance [20]. A more specific and effective classification scheme for serious games also useful for training teams of developers is proposed in [7], that is based upon three mutually independent dimensions: Gameplay, Purpose, and Scope—G/P/S for short. Though originally devised for video games, it can be applied to any type of serious game. A typical example of Gameplay for Agile games is “team-based, non-competitive” (or “team-based, competitive”); the purpose is for instance learning some agile best practice like Test Driven Development or Pair Programming; the scope can target a specific role (e.g. Product Owner), ceremony (e.g. practicing a retrospective) or artifact (e.g. prioritize the product backlog). Given the ubiquity of gamification within the Agile community, it is remarkable that not much research exists on the subject. The most comprehensive study of how serious games are actually used in ICT enterprises is found in [18]. After examining an impressive amount of material, the following groups are found: Serious gaming, Training, Agile methodologies, Future research, Learning theories. There is some research more focused on specific aspects of software development. For instance, [8] examines about 50 games mapping them to the various phases of team development according to Tuckman’s model [19], whereas [13] analyzes how some games can benefit the development process, following specifically the Open Kanban model. A recent work on serious games for teaching agile methods is [15]: this is a systematic literature review to analyze the studies reviewing the use of serious games with the learning objective of teaching Agile Methods for software development. Although serious play activities are very popular in the agile community, there is no complete study about their classification and evaluation concerning team building, role building, and focusing on online play.

3 Team Building for Agile Development One of the most important Agile principles1 is “The best architectures, requirements, and designs emerge from self-organizing teams”. Self-organization is important for 1

https://Agilemanifesto.org/principles.html.

Training Students as Agile Developers: Team and Role …

293

making autonomous the team, and team autonomy is important because making software requires a lot of decisions that should be taken correctly and rapidly in order to avoid delays and waste of time. However, creating a highly performative team is difficult and possibly timeconsuming. While mastering the craft of software development is critical to delivering good software products, many impediments can be overcome supporting specialized communication patterns. Creating an environment that values regular and open interactions, both within the team delivering the product and within the wider stakeholder group can go a long way in removing impediments during the delivery process. Agile team models are well known to reduce specialized roles. In a Scrum team, for instance, only three roles are suggested: Product Owner (a person who has the control of the requirements), ScrumMaster (a person who is in charge of the Scrum process), and developer (the rest of the team is composed by developers). The role of coach is not really part of a Scrum team. The coach has the responsibility to enhance collaboration inside Agile teams, especially when a project involves several teams. Coaches typically use games for a variety of Agile team events, like sprints and release planning sessions [1]. These games help create a shared vision of what the sprint or release will achieve. In a project we have developed for a Public Administration in our country [4], we have been involved in defining a procedure for team building in order to train a group of developers mixing software developers and domain experts, e.g. in cybersecurity. Creating collaborative, high performance teams able to cooperate on problems of digital transformation was a task that required some time, and we employed a number of coaching activities based on games.

4 A Team Building Game for Online Training to Scrum Scrumble2 is a board game based on the roles, events and artifacts of Scrum. We use it to train students with no experience with Scrum or agile. The goal is to show to a team the steps a Scrum team would carry out in order to deploy a new product. The game simulates an experience which passes through some common problems, allowing the game group to test its team building skills facing up many decisions to take. The aim of Scrumble is to complete every User Story required to develop the product, throughout the gradual accomplishment of its specifications during a chosen number of Sprints, before the technical debt becomes unsustainable. The technical debt represents every complication which can shows up in a project and it is related to “immature, incomplete or inadequate artifacts in the software development cycle that cause higher costs and lower quality” : it can be a bug, a testing error, team interaction problems and so on. The first phase of the game consists in the Product Owner presenting to the team the product they will have to create during the game. The better he explains his vision 2

Rules and game material available in http://scrumble.pyxis-tech.com, license is CC-A-NC-ND.

294

P. Ciancarini and M. Missiroli

Fig. 1 A spreadsheet to play Scrumble online

about the product, the more the team will feel involved and interested in playing actively. The PO is in charge of sharing the Product Backlog and decide which User Stories have to be achieved in every sprint. At the beginning, the PO presents every User Story as clearer as he can, in a way the team is able to understand; when this is done, team members have to subjectively estimate the complexity of each story in terms of actual implementation. The first phase of the game consists in the Product Owner presenting to the team the product they will have to create during the game. The better he explains his vision about the product, the more the team will feel involved and interested in playing actively. The PO is in charge of sharing the Product Backlog and decide which User Stories have to be achieved in every sprint. At the beginning, the PO presents every User Story as clearer as he can, in a way the team is able to understand; when this is done, team members have to subjectively estimate the complexity of each story in terms of actual implementation. In order to perform team building during the COVID pandemic we decided to support playing Scrumble in online meetings, hence we transposed the game elements in an Excel sheet, shown in Fig. 1; fast and simple to use, it made the students organize well all the set up needed for the game: the board, the table with the User Stories, the pawns, and the effect cards to pick up during the game. Student teams found useful this game not only to learn Scrum, but also to select the Scrum Master (who has to patiently and masterfully explain the rules) and the Product Owner (who has to master rapidly the art of user story selection and prioritization.

Training Students as Agile Developers: Team and Role …

295

Table 2 A GQM grid for self evaluation by a team after a game with scrumble G1 Learn Do team members understand the Correct answers after a quiz on scrum scrum roles and ceremonies? Do team members feel they learned Opinions from the participants scrum? G2 Practice Was the team able to practice scrum Number of user stories completed effectively? Did team members uniformly Similarity in evaluating the size and estimate during sprint planning? the priority of user stories G3 Harmonize Do team members know each other Number of answers in a questionnaire better? concerning cooperation Does the game let all players to Record and count communication acts cooperate? G4 Motivation Do team members encourage Number of answers in a questionnaire colleagues in need? concerning entertainment Did the PO and the SM help the team? Grading from other players G5 Negotiate Does the team organize their tasks Level of the technical debt at the end properly? of the game Did PO and team plan efficiently the Average of tasks left at the end of sprint backlog? each sprint The first column contains the goals; the second column contains the some questions related to the goal; the third column contains the metrics usable to answer the questions

5 Evaluating Games for Team Building In the evaluation of an activity we need to make clear the goals of the activity itself. The results of any activity should be evaluated also in function of the context in which it is performed. For instance, Lego Serious play can be used in high school, in a university course, or in a business context. These are different situations in which the goals of the teacher/coach proposing and facilitating the game are different, and also different are the evaluation models and purposes. The quality of Agile teamwork is a well known subject of study. For instance, in [11] the effect of collaboration on team performance, problem solving ability, and personal feelings was investigated. In our case we need a framework for evaluating how the gamified teamwork improves team performance. The “Goal-Question-Metric” (GQM) approach is a well known method for driving goal-oriented measures concerning software development. We start by defining the goals we are trying to achieve, then define some questions which require some data to collect. By mapping game outcomes to datadriven metrics, we can form a coherent picture of the serious game activity, and try to understand if we are progressing toward the overall goal of building an Agile, high performing, team. In Table 2 we show the GQM scheme that we developed for the game Scrumble, that is a simulation of Scrum.

296

P. Ciancarini and M. Missiroli

Leveraging the five items defined as pillars of Cooperative Thinking in Table 1, in Table 2 we propose to the students the following five self-evaluation goals: (G1) Learn and expand the knowledge of the Agile process by the team members; (G2) Practice the Agile vision, aiming at improving team members⣙ ability and willingness to help and support each other in carrying out their tasks; (G3) Harmonize the activities and feelings of the team members among themselves; (G4) Fun tries to evaluate if the people involved in the games enjoyed the activities proposed; (G5) Teach the students that any development activity can be negotiated inside the team, evaluating alternatives and sharing the motivations of design choices. We have used a similar schema for gathering data when our students are involved in some team building game: for instance, together with Scrumble we have used Escape the Boom, which can be played over smarthphones. Our goal is to improve their collaboration attitudes, aiming at enhancing reciprocal knowledge and trust inside the team, so that personal differences and abilities are clarified. We used our experiences with these games also in a professional setting, to train team building inside some companies. The procedure we created for team building includes the following steps: 1. assess the current state of the teams, using an Agile maturity model [17], in order to rate the current agility of each team; the model can also be used by a team for self assessments; 2. create a specific unique environment for each team and other stakeholders; the environment should be convenient to host the selected serious games; 3. a whole day, maybe two, should be devoted to team building tasks; 4. the first active phase by the team includes some cooperation games to test the attitude of team members toward cooperative thinking and experimental learning; we have found that the Jenga test game and Magic Maze are especially fun and instructive; 5. organize at least one simulation game in order to show how an Agile development process is structured, with special emphasis on the product owner role; 6. try at least one session of social coding to train programmers in pair and then mob programming; we found that both Refactoring Golf and Code Retreat are quite effective games; 7. try at least one gamification experience with Trello or similar tools, like Taiga; show how to communicate via Slack or Mattermost; exploit Kahoot for Q&A games concerning Agile knowledge; 8. use retrospectives to analyze by GQM the performance of the team during the last development iteration and decide if some team (re)building activity based on games should be performed.

Training Students as Agile Developers: Team and Role …

297

5.1 Validity Threats In our research on exploiting games for team building the internal validity is good, as the data we have from our project-oriented courses show that the students improve their competence of agile cooperation practices after playing some serious games to learn agile. We have searched references for games proposed for team building in a systematic way, analyzing papers in agile conferences and sites specialized in coaching agile teams. The external validity concerning the generalization of the results of our approach based on GQM has been confirmed because when trying new games our students were able to apply effectively the GQM approach to evaluate their own progress. The construct validity is moderately good and difficult to improve because the available literature is scarce: most games have been published informally and sometimes the rules and the prospected team interactions are contradictory. The main limitation of our work is that the team building activities have the goal to improve the the reciprocal understanding of people with some experience and a good understanding of software development. We have seen that especially more expert people appreciate the games, while computer science students are more entertained than convinced. Another important limitation is that all games were played in presence, with the notable exception of coding games which were supported by co-editing tools. We also made no experience with people with physical handicaps.

6 Conclusions and Future Works We have studied the topic of team building in the context of a contract for agile developments of software services for a public administration. We have surveyed several team building games especially invented for training developers in agile frameworks. For Scrum, we have found quite effective the game Scrumble, that we have adapted to an online version. We have developed a GQM evaluation grid based on the main concepts developed by Cooperative Thinking. We made several experiments with both our students of software development classes and some professionals in specific training workshops. We have found that team building activities can be successfully organized as team games, and we have experimented a number of these games with developers of all ages, who enjoyed the experiments and declared to have improved their competency concerning agile software development practices (e.g. pair programming swapping the pairs during an iteration). Before the pandemic we were able to make a number of experiments with people face-to-face, asking for their impressions at the end; we got excellent ratings and encouraging remarks. The advent of pandemic compelled us to switch to remote communication for our activities, so the game list had to be revised and new tools introduced, in particular audio/video platforms like Microsoft Teams and Zoom. We are now engaged in experiments of remote play with some student teams, either in

298

P. Ciancarini and M. Missiroli

some high schools or in our university; we are collecting data useful to refine our team building model. The results of these experiments will by the topic of a future work.

References 1. Adkins, L.: Coaching Agile Teams: A Companion for Scrum Masters, Agile Coaches, and Project Managers in Transition. Addison-Wesley (2010) 2. Caillois, R.: Man, Play, and Games. University of Illinois Press (2001) 3. Ciancarini, P., Missiroli, M., Russo, D.: Cooperative thinking: analyzing a new framework for software engineering education. J. Syst. Softw. 157 (2019) 4. Ciancarini, P., Missiroli, M., Russo, D.: Exploiting Agile practices to teach computational thinking. In: Bruel, J.M., Mazzara, M., Meyer, B. (eds.) Software Engineering Aspects of Continuous Development and New Paradigms of Software Production and Deployment. Lecture Notes in Computer Science, vol. 12055, pp. 63–83. Springer, Berlin (2020) 5. Conway, M.: How do committees invent. Datamation 14(4), 28–31 (1968) 6. Denning, P.J., Tedre, M.: Computational Thinking. MIT Press (2019) 7. Djaouti, D., Alvarez, J., Jessel, J.P.: Classifying serious games: the G/P/S model. In: Handbook of Research on Improving Learning and Motivation Through Educational Games: Multidisciplinary Approaches, pp. 118–136. IGI Global (2011) 8. Jovanovi´c, M., Mesquida, A.L., Radakovi´c, N., Mas, A.: Agile retrospective games for different team development phases. J. Univers. Comput. Sci. 22(12), 1489–1508 (2016) 9. Katzenbach, J.R., Smith, D.K.: The Wisdom of Teams: Creating the High-Performance Organization. Harvard Business Review Press (2015) 10. Korkmaz, O., Cakir, R., Yasar Ozden, M.: A validity and reliability study of the computational thinking scales (CTS). Comput. Hum. Behav. 72, 558–569 (2017) 11. Lindsjørn, Y., Sjøberg, D.I., Dingsøyr, T., Bergersen, G.R., Dybå, T.: Teamwork quality and project success in software development: a survey of agile development teams. J. Syst. Softw. 122, 274–286 (2016) 12. Missiroli, M., Russo, D., Ciancarini, P.: Learning Agile software development in high school: an investigation. In: Proceedings of the International Conference on Software Engineering, pp. 293–302. ACM/IEEE (2016) 13. Przybylek, A., Olszewski, M.K.: Adopting collaborative games into open kanban. In: 2016 Federated Conference on Computer Science and Information Systems (FedCSIS), pp. 1539– 1543. IEEE (2016) 14. Przybyłek, A., Zakrzewski, M.: Adopting collaborative games into agile software development. In: Proceedings of the 13th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE 2018). Communications in Computer and Information Science, vol. 1023, pp. 119–136. Springer, Berlin (2018) 15. Rodríguez, G., González-Caino, P.C., Resett, S.: Serious games for teaching agile methods: a review of multivocal literature. Comput. Appl. Eng. Educ. 29(6), 1931–1949 (2021) 16. Russo, D., Stol, K.J.: PLS-SEM for software engineering research: an introduction and survey. ACM Comput. Surv. (CSUR) 54(4), 1–38 (2021) 17. Schweigert, T., Vohwinkel, D., Korsaa, M., Nevalainen, R., Biro, M.: Agile maturity model: analysing agile maturity characteristics from the SPICE perspective. J. Softw. Evol. Process 26(5), 513–520 (2014) 18. Stettina, C.J., Offerman, T., De Mooij, B., Sidhu, I.: Gaming for agility: using serious games to enable agile project & portfolio management capabilities in practice. In: 2018 IEEE International Conference on Engineering, Technology and Innovation (ICE/ITMC), pp. 1–9. IEEE (2018)

Training Students as Agile Developers: Team and Role …

299

19. Tuckman, B.W.: Developmental sequence in small groups. Psychol. Bull. 63, 384–399 (1965) 20. Zuppiroli, S., Ciancarini, P., Gabbrielli, M.: A role-playing game for a software engineering lab: developing a product line. In: Proceedings of the 25th IEEE Conference on Software Engineering Education and Training (CSEET), pp. 13–22, Nanjing, China (2012)

Impact of Covid-19 on Employee Satisfaction and Trust with Focus on Working from Home Miriam Gazem, Ralf Härting, Anna Schneider, and Christopher Reichstein

Abstract Due to the Covid-19 pandemic, companies were forced to send employees into home office. This presented major challenges for many companies. In addition, this meant a change of habits, both for employees and managers. The paper focuses on the question how this circumstance has an influence on employee satisfaction. Therefore, a theoretically based empirical study was conducted. A conceptual model was created, and hypotheses were formulated for this purpose based on a structured literature review. Isolation, communication, and trust (from supervisors’ and employees’ perspective) were identified as influencing factors that were examined in this study. It could be shown that the biggest influencing factors are isolation and supervisor trust. The influence of communication on employee satisfaction was confirmed, too. For an outlook, linear regression was conducted.

1 Introduction At the beginning of the Covid-19 pandemic a combat attempt was made by means of a lockdown. As a result, companies have sent their employees into home office. Research from the second quarter of 2020 show that the number of people working from home is up to 61% of the workforce. The potential ranges up to 80%. The use of home office to escape the pandemic is reflected in the increase of the usage rate from 39 to 61% [1]. Since the number of studies on this topic in Germany is still limited, the study focuses exclusively on employees and supervisors in Germany. This study examines whether employee satisfaction changes in home office. There are various components that have an influence on employee satisfaction. This M. Gazem · R. Härting (B) · A. Schneider Aalen University of Applied Science, Beethovenstraße 1, 73430 Aalen, Germany e-mail: [email protected] C. Reichstein Cooperative State University, Baden-Württemberg, Marienstraße 20, 89518 Heidenheim, Germany © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_27

301

302

M. Gazem et al.

study will focus on four main influencing factors “isolation”, “communication”, “supervisor trust” and “employee trust”.

2 Theoretical Background The present study was preceded by a literature review. None of the available studies refer explicitly to how the Covid-19 pandemic and the resulting increased occurrence of home offices affect employee satisfaction. In the present study the already known influential constructs “communication” and “isolation” are applied to the challenging pandemic situation and are supplemented by the construct “trust in home office”. Employee satisfaction. It has already been proven that home offices have an impact on employee satisfaction [3]. In the past there has already been many studies about the influence of teleworking on employee satisfaction [4]. In this context employee satisfaction is considered one of the most frequently studied topics [5–9]. To investigate the impact of home office on employee satisfaction this research assumes that employee satisfaction is conditioned by employee motivation. Motivation was operationalized by Herzberg’s hygiene factors as well as by intrinsic and extrinsic motivation [10]. For employees to be motivated the hygiene factors must be fulfilled. Intrinsic and extrinsic motivation derive from the self-determination theory of Deci and Ryan [11]. Interest-driven actions are perceived as intrinsically motivated behaviors [11, 12]. Additionally, there is extrinsic motivation. This is often based on a request that leads to the expectation of a consequence [11]. The extrinsic motivation was divided into external, introjected, identified and the integrated regulation [13–15]. Isolation. Home office employees are less likely to identify with their team, less likely to share experiences, less likely to receive feedback and more likely to be ostracized [16, 17]. They are also less able to relate to others. This leads to a change in affective relationships and can have negative effects on motivation, engagement, and work performance [18, 29]. The reduced emotional involvement in home office can be both an advantage and a disadvantage [19]. Social isolation in home office can be reduced by receiving more support from colleagues and superiors [20]. Separation from colleagues can also lead to dissatisfaction [21]. Constant work in the home office can result in a reduction of opportunities for further development [22]. The negative effects are intensified if the manager is also working from home [23]. The most cited study focuses on the impact of isolation on work performance and job change intentions [24]. The constructs of home office and isolation are present in all the cited references, but the circumstances may change significantly in the context of a global pandemic. This paper attempts to relate the construct of isolation to employee satisfaction much more clearly than in previous work rather than focusing on pure job performance

Impact of Covid-19 on Employee Satisfaction and Trust with Focus …

303

or turnover intentions. To operationalize isolation, the individual feeling of loneliness and the need for belonging were considered [25, 26]. The hypothesis to be investigated is: H1: “Home office isolation has an effect on employee satisfaction.” Trust. Trust in its definition is usually considered as a feeling, conviction or an expectation towards a partner resulting from reliability, competence, and purposefulness [27]. Mayer et al. describe trust in an organization in their “Proposed Model of Trust”. The trusting party (trustor) and a party to be trusted (trustee) are involved. The qualities of a trustworthy person include ability, benevolence, and integrity [28]. The study “Examining the effects of trust in leaders: A bases-and-foci approach distinguishes the two types of trust (affective and cognitive) [33]. The study concludes that employee satisfaction can result from cognitive trust in management and affective trust in the direct supervisor and a distinction between these types should occur. Trust has a direct impact on job satisfaction as it is defined as the willingness to be vulnerable [28–33]. Based on the findings it appears that the distinction between cognitive and affective trust, ability, benevolence, and integrity are important for further study, so this distinction is included in this study. It asks whether trust between supervisors and their employees influences employee satisfaction in the home office. The following hypotheses are formed: H2: “The trust of the employee in the supervisor in the home office has an impact on employee satisfaction.” H3: “The trust of the supervisor in the employees in the home office has an impact on employee satisfaction.” Communication. In the context of communication at home office aspects of the Communication Satisfaction Questionnaire (CSQ) by Downs and Hazen (1977) and the associated eight dimensions of communication satisfaction are included [34]. The extent to which communication leads employees to identify with the organization is included in the communication climate dimension. The supervisor’s attention to the employees, the offer to guide in case of professional problems and the openness to ideas can be assigned to supervisory communication. Preservation of information from the work environment is addressed in the dimension of organizational integration. Assessing the performance of employees and evaluating them are part of personal feedback. To discuss whether informal communication is fluent and accurate, the dimension of co-worker communication is considered. Information that relates to the entirety of the organization such as goals of an organization is referred to as corporate information. Upward and downward communication with subordinates is included in the Subordinate communication [34]. For this reason, the following hypothesis is formulated.

304

M. Gazem et al.

H4: “Communication in the home office has an influence on employee satisfaction.” The present study applies the presented constructs to the completely changed situation due to the global pandemic in Germany.

3 Research Methods and Data Collection The conceptual model underlying the study is based on the structured literature review and is shown in Fig. 1. A deductive approach was taken. This study is limited to the influence of “isolation”, “communication” and “trust”. Trust is considered separately from the perspective of supervisors and employees. The study was examined with a cross-sectional quantitative research approach using a web-based survey (LimeSurvey) that is derived from the research model. The questionnaire can be divided into five areas of sample description and the constructs of employee satisfaction, isolation, trust, and communication [35]. A pretest was conducted to check the structure as well as the consistency of the questionnaire. The analysis was performed with SPSS version 27.0.1.0. Cronbach’s alpha was used to test the quality of the questionnaire [36, 37]. The main study includes a sample of n = 424 responses. All 83 items were rated by participants using a 6-point scale (1 = fully agree, 6 = do not agree at all). After cleaning the data, a final sample of n = 259 employees and supervisors was obtained. Employed managers as well as employees from all industries and company sizes were surveyed.

Fig. 1 Conceptual model

Impact of Covid-19 on Employee Satisfaction and Trust with Focus …

305

In the first part of the questionnaire all descriptive aspects (20 items) are discussed. About 45% of the participants were female, 54% were male and 1% answered that they belong to the gender diverse. Large companies with more than 250 employees use 83% of participants. 6% of the participants come from a medium-sized company with up to 250 employees. From small companies with less than 50 employees, 11% of respondents participated in the study. The questionnaire discussed whether respondents are currently on short time due to the Covid-19 pandemic. 67% (174 people) of respondents said they were not on short-time work. 36% (92 people) of participants had lost salary at the time of the survey due to the Covid-19 pandemic. 51% of the participants (133 people) reported that they can work in a home office only since the Covid-19 pandemic. The next four parts of the survey consist of the individual constructs. To determine the reliability of the questionnaire the Cronbach’s alpha values of the constructs were formed and the selectivity of the items was considered. Items that had a selectivity of less than 0.3, were excluded from the analysis [38]. According to Nunnally, the minimum value of the Cronbach’s alpha of 0.7 is achieved and the constructs can be considered acceptable [40].

4 Results Descriptive statistics. In Table 1, the mean values, standard deviations, and Cronbach’s alpha values for the individual constructs can be viewed. Regarding the item on whether home office takes away private retreat space, 28.5% of the respondents confirmed this statement. 71.5% of respondents indicated no interference with private space. The respondents were also asked whether they felt that they needed to be constantly available in their home office 35.1% of respondents stated that they feel they must be available around the clock during home office. 23.5% of the respondents stated that they tended to feel a double load when working from home. 76.4% tended to reject the statement. Men feel more often double burdened with 28.78%. Among women, this concerns 16.24%. Table 1 Results Sample size n

Mean value

Standard deviation

Variance

Cronbach’s alpha

Employee satisfaction

259

2.55

0.71

0.51

0.759

Isolation

259

3.07

0.85

0.72

0.87

Communication

259

2.32

0.74

0.55

0.923

Supervisor trust

59

1.69

0.52

0.27

0.908

Employee trust

200

1.97

0.82

0.68

0.931

306

M. Gazem et al.

Another item included home office productivity. 40.9% of respondents said they were just as productive at home as in the office. 40.2% of respondents said they were more productive than in the office and 18.9% said they were less productive. Regarding the frequency of exchange 47.1% of the respondents stated that they exchange with their team colleagues several times a day. The exchange once a day within the team was reported as 20.1%. Besides 23.2% of the respondents stated that they exchange with their colleagues several times a week. 8.9% of the respondents stated that they exchange with their colleagues once a week. Regarding the frequency of communication with their supervisor 13.9% of respondents said that they communicate with their supervisor several times a day. Additionally, 17.4% of the respondents communicate with their supervisor once a day. 38.6% indicated communicating with their supervisor several times a week. Once a week was indicated by 20.5%. Lastly, 9.7% reported communicating with their supervisor less than once a week. It was established that 76.3% of the supervisors said they control their employees just as often in the home office as in the office. A more frequent control in home office than in the office was indicated by supervisors with 5.1%. The result of controlling employees during home office less frequently than in the office amounts to 18.6%. From the employees’ point of view 64.5% surveyed they felt controlled just as often as in the office. 5.5% said they felt controlled more often than in the office. Feeling less frequently controlled in the home office compared to the office was reported by 30%. Regression analyses. The null hypotheses to be investigated state that there is no influence of the considered constructs on employee satisfaction. To test the hypotheses a multiple regression analysis was conducted. This allows the constructs “isolation” and “communication” to be considered in one analysis. For trust the sample was divided into employees and supervisors. A linear regression analysis was performed for each. Since multiple regression examines several constructs at once the probability of making false statements increases. For this reason, the alpha level for the multiple regression was adjusted to the value 0.05/2 = 0.025 using a Bonferroni correction. In this way, the confidence interval is 97.5% [40]. Four outliers were excluded, homoscedasticity was determined, and autocorrelations were excluded using the Durbin-Watson statistic [34, 36]. There is no multicollinearity according to the Pearson correlation [37]. Then a Kolmogorov–Smirnov test was performed. This resulted that the constructs “isolation”, “employee satisfaction” and “supervisors’ trust” are normally distributed only “employee trust” is not [38]. A linear regression analysis for employee trust was performed for the outlook. For the multiple regression analysis of the two constructs the R-value is R = 0.703. A strong correlation is assumed for values above 0.5 [38]. The coefficient of determination R2 = 0.494 indicates that the examined constructs predict 49.4% of the dispersion of employee satisfaction. There is a variance explanation of 49.4% [40]. To determine the effect size Cohen’s method is used: F2 = R2 /(1 − R2 ). This results in a value of 0.976 for F2 . The effect strength is considered to be strong from an F2 value of 0.35 upwards. With an F2 of 0.976, the model has a large effect strength [39]. The

Impact of Covid-19 on Employee Satisfaction and Trust with Focus …

307

significance level is set at 2.5%. The predictors “isolation” and “communication” statistically significantly predict the criterion “employee satisfaction”, F (2,256) = 83.352, p < 0.001. The model is transferable to the population [41]. The beta for the construct isolation is 0.606. The significance is < 0.001. For communication, the beta is 0.136. The significance is 0.022, which is still below the required 0.025 [42]. Employee satis f action = 0.508 ∗ isolation + 0.131 ∗ communication − 0.701 (constant) Conclusion for communication and isolation. If communication increases by one unit employee satisfaction in home office increases by 0.131. If isolation increases by one unit employee satisfaction in home office increases by 0.508. For the linear regression of the construct “supervisor trust” the R value is R = 0.496. From 0.3 to 0.5, a medium/moderate correlation is assumed [38]. The coefficient of determination R2 = 0.246 indicates that the examined construct is 24.6% predictive of the dispersion of employee satisfaction [40]. With an F2 of 0.326 the effect strength after Cohan [38] is medium. The significance level is set at 5%. The predictors “supervisor trust” statistically significantly predicts the criterion “employee satisfaction”, F (1,57) = 18.550, p < 0.001 [42]. The beta of the construct “supervisor trust” is 0.496. The significance is < 0.001. The model is transferable to the population [42]. Employee satis f action = 0.677 ∗ Super visor tr ust − 1.439 (constant). Conclusion for supervisor trust. If supervisor trust increases by one unit employee satisfaction increases by 0.677. The linear regression of the construct “employee trust” is only performed for an outlook as the data in this construct are not normally distributed. The R-value for this model is R = 0.301. Values from 0.3 to 0.5 are assumed to have a medium/moderate correlation [40]. The coefficient of determination R2 = 0.090 indicates that the examined constructs predict employee satisfaction at 9% of the variance. There is a variance explanation of 9% [42]. With an F2 of 0.0989, the model has a low effect strength [2]. The significance level is set at 5%. Even if the construct is not normally distributed the predictor “employee trust” statistically significantly predicts the criterion “employee satisfaction”, F (1,198) = 19.700, p < 0.001. The beta of the construct “employee trust” is 0.301. The significance is < 0.001. Employee satis f action = 0.261 ∗ Employee tr ust − 2.029(constant) Conclusion for employee trust. According to the available calculations an increase in employee trust could have a positive effect on employee satisfaction. The value may differ from the value calculated here because the construct does not meet all the requirements of multiple regression.

308

M. Gazem et al.

Fig. 2 Tested model

According to the calculations the biggest factors influencing employee satisfaction are the constructs “supervisor trust” and “isolation” (Fig. 2). The tested null hypotheses can all be rejected. All examined constructs have an influence on employee satisfaction in the home office during the Covid-19 pandemic.

5 Discussion The present results confirm that the constructs “isolation”, “trust” and “communication” have an influence on employee satisfaction. This study demonstrates the positive influence of the reviewed constructs “communication” and “isolation” on employee satisfaction in the home office during the Covid-19 pandemic. These results support the findings of other studies conducted without the influence of the crisis event. An important positive influencing factor was isolation where a rather negative influence was expected during the current situation. Satisfaction is reinforced by the feeling of being able to work in isolation and alone. The initial perspective is that isolation has a positive effect on satisfaction because it leads to fewer interruptions to work due to disturbances from colleagues and superiors. The greatest positive influence on communication satisfaction was the availability of the manager and personal identification with the company. The availability of the manager should therefore be ensured as a matter of principle. Personal identification with the company also addresses an important motivational factor. The goal should

Impact of Covid-19 on Employee Satisfaction and Trust with Focus …

309

be to retain employees in the home office as well. In this context it should be noted that employees should feel taken seriously by their team and their manager. In addition, a positive effect of trust both from supervisors’ as well as employees’ perspective on employee’s satisfaction can be assumed. According to the results, building trust and maintaining it on an ongoing basis is an important management task that is significant for employee satisfaction in the home office. The Covid-19 crisis created a new set of challenges for companies and their employees. This study showed that the influence of the examined constructs on employee satisfaction also applies to home office during the crisis. It could be proven that there is an influence between satisfaction in the home office and the trust relationship between manager and employee. The main challenges seem to arise on an emotional level. There is a reason to believe that empathy plays a particularly important role during such a time. Physical resources or workshop are not available for strengthening the team and company identification. The management culture has to redefine itself according to the current challenges. Recommendation. Home office in the Covid-19 pandemic has presented major challenges to many companies. To strengthen communication satisfaction the availability of the manager should always be ensured. The information provided to employees should not be reduced compared to face-to-face work. In addition to regular jour fixes and task-related communication via e-mail, telephone or video call the exchange of information which otherwise tends to take place from door to door in the office, should be covered by technical aids. Even in everyday office life not all information has the same priority, but it can be important for the perceived involvement. These can be used for communication between managers and employees as well as for communication within the team and ensure perceived closeness to teammates. Such an offering could be supplemented by granting the use of the company’s internal communication tools for joint, digital break activities such as discussion rounds or games. Supervisors should be sensitized through coaching to ensure that communication satisfaction is considered in the virtual management of their team. The feeling of isolation during home office has a fundamentally positive effect on employee satisfaction in the home office. The option of working from home should be retained even after the Covid-19 crisis. In addition, employees should have the option of being able to isolate themselves in the office as well. Trust has an impact on employee satisfaction. In the case of supervisor trust the confidence of the tasks and the perceived honesty play a major role. Differences in the requirements and skills of employees should be quickly identified so that rapid, digital training measures or adjustments in the distribution of tasks can be initiated. Personal trust between supervisor and employee is therefore a complex construct as it is based on personal experience and is built up slowly. Basically, the measures here depend heavily on the existing team culture. Supervisors should consider strengthening the trust of their employees as an important part of their job especially in times of the Covid-19 pandemic. Limitations. The model utilized in this study makes an important contribution to elucidating the fundamental impact of home office during the Covid-19 pandemic

310

M. Gazem et al.

on employee satisfaction. However, this does not claim to be exhaustive. It is likely that there are other constructs that have an impact on home office satisfaction during the Covid-19 crisis. For example, the constructs “stress” and “anxiety” could be mentioned which should be investigated in further studies. Examples of fears that could arise in conjunction with the Covid-19 pandemic are basically fear of contracting the virus or fear of losing a job due to the worsened economic situation. Acknowledgements We thank Alina Gehrig, Demian Deffner and Chiara Frank for supporting our research. Thank you for your valuable and helpful contributions.

References 1. Ifo. https://www.ifo.de/personalleiterbefragung/202008-q2. Last accessed 2021/02/06 2. Vaziri, H., Casper, W.J., Wayne, J.H., Matthews, R.A.: Changes to the work-family interface during the COVID-19 pandemic: examining predictors and implications using LTA. J. Appl. Psychol. 105(10), 1073–1087 (2020) 3. Smith, S.A., Patmos, A., Pitts, M.J.: Communication and teleworking: a study of communication channel satisfaction, personality, and job satisfaction for teleworking employees. Int. J. Bus. Commun. 55(1), 44–68 (2018) 4. Gajendran, R.S., Harrison, D.A.: The good, the bad, and the unknown about telecommuting: meta-analysis of psychological mediators and individual consequences. J. Appl. Psychol. 92(6), 1524–1541 (2007) 5. Golden, D.T.: The role of relationships in understanding telecommuter satisfaction. J. Organ. Behav. 27, 319–340 (2006) 6. Cooper, C.D., Kurland, N.B.: Telecommuting, professional isolation, and employee development in public and private organizations. J. Organ. Behav. 23(4), 511–532 (2002) 7. Golden, D.T., Veiga, F.J.: The impact of extent of telecommuting on job satisfaction: resolving inconsistent findings. J. Manage. 31(301), 301–318 (2005) 8. Vega, R.P., Anderson, A.J., Kaplan, S.A.: A within-person examination of the effects of telework. J. Bus. Psychol. 30(2), 313–323 (2015) 9. Fonner, K.L., Roloff, M.E.: Why teleworkers are more satisfied with their jobs than are officebased workers: when less contact is beneficial. J. Appl. Commun. Res. 38(4), 336–361 (2010) 10. Herzberg, F.: One more time. How do you motivate employees? Harvard Bus. Rev. 5–16 (1987) 11. Deci, E.L.: Interest and the intrinsic motivation of behavior. In: Renninger, K.A., Hidi, S., Krapp, A. (eds.) The Role of Interest in Learning and Development, pp. 43–70. Hillsdale, NJ 1992 (2005) 12. Deci, E.L., Ryan R.M.: Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. Am. Psychol. 55, 68–78 (2000) quoted from Deci, E.L.: Intrinsic Motivation. New York (1975) 13. Deci, E.L., Ryan, R.M.: Intrinsic Motivation and Self-determination in Human Behavior. New York (1985) quoted from Deci, E.L., Ryan, R.M.: Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. Am. Psychol. 55, 68–78 (2000) 14. Deci, E.L.: Effects of externally mediated rewards on intrinsic motivation. J. Pers. Soc. Psychol. 18, 105–115 (1971) 15. Deci, E.L.: Intrinsic motivation, extrinsic reinforcement, and inequity. J. Pers. Soc. Psychol. 22, 113–120 (1972) 16. Vega, G., Brennan, L.: Isolation and technology: the human disconnect. J. Organ. Change Manage. 13(5), 468–481 (2000)

Impact of Covid-19 on Employee Satisfaction and Trust with Focus …

311

17. Allen, T.D., Golden, T.D., Shockley, K.M.: How effective is telecommuting? Assessing the status of our scientific findings. Psychol. Sci. Public Interest 16(2), 40–68 (2015) 18. Mann, S., Varey, R., Button, W.: An exploration of the emotional impact of tele-working via computer-mediated communication. J. Manag. Psychol. 15, 668–690 (2015) 19. Ashfort, B.E., Humphrey, R.H.: Emotion in the workplace a reappreasial. Human Relat. 48(2), 97–125 (1995) 20. Bentley, T.A., Teo, S.T.T., McLeod, L., Tan, F., Bosua, M., Floet, M.: The role of organisational support in teleworker wellbeing: a sociotechnical systems approach. Appl. Ergon. 52, 207–215 (2016) 21. Boel, S.K., Cecez-Kecmanovic, D., Campbell, J.: Telework paradoxes and practices: the importance of the nature of work. New Technol. Work Employ. 31(2), 114–131 (2016) 22. Cooper, C.D., Kurland, N.B.: Telecommuting, professional isolation, and employee development in public and private organizations. J. Organ. Behav. 23, 511–532 (2002) 23. Park, S., Cho, Y.J.: Does telework status affect the behavior and perception of supervisors? Examining task behavior and perception in the telework context. Int. J. Human Resour. Manage. https://doi.org/10.1080/09585192.2020.1777183. Last accessed 2020/11/21 24. Golden, D.T., Veiga, J.F., Dino, R.N.: The impact of professional isolation on teleworker job performance and turnover intentions: does time spent teleworking, interacting face-to-face, or having access to communication-enhancing technology matter? J. Appl. Psychol. 93(6), 1412–1421 (2008) 25. Baumeister, R.F., Tice, D.M.: Anxiety and social exclusion. J. Soc. Clin. Psychol. 9(2), 165–195 (1990) 26. Baumeister, R., Leary, M.: The need to belong: desire for interpersonal attachments as a fundamental human motivation. Psychol. Bull. 117(3), 497–529 (1995) 27. Moorman, C., Zaltman, G., Deshpande, R.: Relationships between providers and users of market research: the dynamics of trust within and between organizations. J. Mark. Res. 29, 314–328 (1992) 28. Mayer, R., Davis, J., Schoorman, D.: An integrative model of organizational trust. Acad. Manag. Rev. 3, 709–734 (1995) 29. Pillai, R., Schriesheim, C.A., Williams, E.S.: Fairness perceptions and trust as mediators for transformational and transactional leadership: a two-sample study. J. Manag. 25, 897–933 (1999) 30. Rotter, J.: A new scale for the measurement of interpersonal trust. J. Pers. 35(4), 651–665 (1967) 31. McAllister, D.J.: Affect- and cognition-based trust as foundations for interpersonal cooperation in organizations. Acad. Manage. J. 38, 24–59 (1995) 32. Schoorman, F.D., Mayer, R.C., Davis, J.H.: An integrative model of organizational trust: past, present, and future. Acad. Manag. Rev. 32, 344–354 (2007) 33. Yang, J., Mossholder, K.: Examining the effects of trust in leaders: a bases-and-foci approach. Leadersh. Q. 21, 50–63 (2010) 34. Clampitt, P.G., Downs, C.W.: Employee perceptions of the relationship between communication and productivity: a field study. J. Bus. Commun. 7–9, 13–15 (1993) 35. Questionnaire and further results. http://www.kmu-aalen.de/kmu-aalen/appendix.pdf 36. Cohan, J., Cohan, P., West, S.G., Aiken, L.S.: Applied Multiple Regression/Correlation Analysis for the Behavioral Science, 3rd ed., pp. 64–99 (2003) 37. Pearson, K.: On a new method of determining correlation between a measured Character A, and a Character B, of which only the percentage of cases wherein B exceeds (or falls short of) a given intensity is recorded for each grade of A. Biometrika 7(1/2), 96–105 (1909) 38. Massey, F.J.: The Kolmogorov-Smirnov test for goodness of fit. J. Am. Stat. Assoc. 46(253), 68–78 (1951) 39. Cohen, J.: A power primer. Psychol. Bull. 112(1), 155–159 (1992) 40. Lovel, M.C., Prescot, E.: Multiple regression with inequality constraints: pretesting bias, hypothesis testing and efficient. J. Am. Stat. Assoc. 65(330), 913–925

312

M. Gazem et al.

41. Wasserstein, R.L., Lazar, N.A.: The ASA statement on p-values: context, process, and purpose. Am. Stat. 70(2), 129–133 (2016) 42. Mason, C.H., Perreault, W.D.: Collinearity, power, and interpretation of multiple regression analysis. J. Mark. Res. 28(3), 268–280 (1991)

An Approach for Organising and Managing an Academic Year Using Online Tools and Techniques Liviu-Andrei Scutelnicu and Marius Ciprian Ceobanu

Abstract E-learning is a concept in continuous research, which provides for the instruction of pupils/students/trainees in a dynamic way, incorporating certain aspects from all forms of modern education. It is desired both to increase the knowledge of the individual and to increase the efficiency of interpersonal relationships, so much affected by SarsCov2. In the context of the pandemic, in order to continue working with students, we chose to organise and manage the academic year in a more special way, namely, to use one of the free tools available on the Internet, but with the possibility to customise it according to our needs and requirements. In this paper we will present a way in which we managed the academic year, applied to the Faculty of Computer Science, from our university and what additional development we took into account to succeed in adapting the program to our needs. We took into consideration the Discord platform, first because it is an open-source (free) platform and second because it has the possibility and offers the opportunity to implement socalled bots, which are in fact programs implemented by developers, which perform certain processes automatically or semi-automatically, with certain pre-established conditions at the time of development. The purpose of this study is to help nontechnical people as well to use this kind of platform for organising and managing their classes, to have an active interaction with their students, even if such events occur that force us to socially distance ourselves.

L.-A. Scutelnicu (B) Faculty of Computer Science, “Alexandru Ioan Cuza” University of Iasi, Iasi, Romania e-mail: [email protected] M. C. Ceobanu Faculty of Psychology and Educational Sciences, “Al. I. Cuza” University of Iasi, Iasi, Romania e-mail: [email protected] L.-A. Scutelnicu Institute for Computer Science, Romanian Academy, Iasi Branch, Bucharest, Romania © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_28

313

314

L.-A. Scutelnicu and M. C. Ceobanu

1 Introduction In the field of artificial intelligence and machine learning [1], we noticed an accelerated evolution that has behind it the huge investments made by states in research, but especially the investments of the big international corporations Google, Apple, Microsoft. These companies have at least three competitive advantages. First advantage is related to the ability to attract the best minds from the academy to their research centres. Second advantage is the availability of model drive data. Another advantage is the easy and sometimes virtually free access to cloud resources for model training and experimentation due to the fact that they can use the resources they do not rent at a given time. Other important advantage of this approach is that these large companies seek to make scientific results available to the whole world (for pragmatic reasons related to the recruitment capacity of researchers) but at the same time on the operating side they are only interested in results that impact billions of people. E-learning [2] is a concept in continuous research, which provides for the instruction of pupils/students/trainees in a dynamic way, incorporating certain aspects from all forms of modern education. It is desired both to increase the knowledge of the individual and to increase the efficiency of interpersonal relationships, so much affected by SarsCov2 [3]. The aim is to move from the first e-learning concepts that aimed only at the use of new technologies to the humanization of the learning process. In the context of the pandemic, in order to continue working with students, we chose to organise and manage the academic year in a more special way, namely, to use one of the free tools available on the Internet, but with the possibility to customise it according to our needs and requirements. In this paper we will present a way in which we managed the academic year, applied to the Faculty of Computer Science, from our university and what additional development we took into account to succeed in adapting the program to our needs. In this study, we will present the development and the applicability of such a program which interacts with the Discord platform and how this program can be reused, to generate such a session, regardless of specialisation, language or even without the need for certain studies in the field of computer science. The user interaction being minimal only by specifying certain information about the session (name of the session, number of groups, session etc.). The program automatically configures and initialises a session on Discord, based on certain criteria such as the user roles (teacher, student, staff etc.) the year of study or by the master’s specialisation (1st year, 2nd year, 3rd year, 1st master—1st specialisation, 1st master—2nd specialisation a.s.o.).

An Approach for Organising and Managing an Academic Year Using …

315

2 Video and Text Streaming Platforms The most used streaming platforms in the world are also those owned by top companies from the IT industry [4], such as Cisco Webex, Microsoft Teams, Google Meet, Zoom, Skype, a.s.o. They all do approximately the same thing, but have certain limitations, limitations that can be removed through various subscriptions paid annually or monthly, depending on each individual provider.

2.1 Zoom Zoom is one of the most used platforms in the area of video streaming, and is owned by Zoom Video Communications (Zoom), which is an American communications technology company headquartered in California. It offers video telephony and online chat services through a peer-to-peer software platform [5] based on cloud and is used for teleconferencing, telecommunications, distance education and social chats. Zoom’s business strategy focuses on providing a product that is easier to use than competitors, as well as cost savings, which include minimizing infrastructure computing costs and ensuring a high degree of employee efficiency. Since 2020, the use of Zoom software has seen a significant global increase due to the introduction of quarantine measures adopted in response to the COVID-19 pandemic. Its software products have faced public and media scrutiny over security and privacy concerns. Some of Zoom’s workforce is based in China, which has given rise to surveillance and censorship issues. Key features of Zoom include: • • • • • •

video chat and HD conferences; audio conferencing using VoIP [6]; court messenger; virtual backgrounds for video calls; screen sharing and collaborative whiteboards; hosting video webinars.

2.2 Cisco Webex Cisco Webex is an American company that develops and sells web conferencing and video conferencing applications. It was founded as WebEx in 1995 and taken over by Cisco Systems in 2007. The most relevant software products are Webex Meetings, Webex Teams, Training Center, Event Center. All Webex products are part of the Cisco Systems collaboration portfolio.

316

L.-A. Scutelnicu and M. C. Ceobanu

In May 2020, Cisco reported that they had 500 million meeting participants, which equates to 25 billion meeting minutes, using its WebEx video conferencing application. In September 2020, Cisco launched a new Webex Classrooms platform for virtual home meetings. The Cisco Webex application is a real-time HD video conferencing platform that users can share with multiple people in the group. One of the advantages of this system is the user access through the web page by installing the plug-in only in the web browser. In the meeting room (or classroom), users can exchange group data files, share presentations. In addition, WebEx is a secure online conference and can also record the meeting for later sharing on the website. The system can be accessed with different devices such as computers, smart phones, tablets, etc. (iOS, Android). In one study [7], a classroom can support up to 100 users. Each user can join 2 groups simultaneously and has the ability to display video conferences with up to 7 participants.

2.3 Microsoft Teams Microsoft Teams is a proprietary business communication platform developed by Microsoft as part of the Microsoft 365 family of products. Teams primarily competes with the similar service Slack, offering workspace chat and video conferencing, file storage, and application integration. Teams replaces other messaging and collaboration platforms operated by Microsoft, including Skype for Business and Microsoft Classroom. Microsoft announced Teams at an event in New York and launched the service worldwide in 2017. It was created during an internal hackathon at the company’s headquarters and is currently led by Microsoft corporate vice president Brian MacDonald. Teams allow communities, groups, or teams to join via a specific URL or an invitation sent by an admin or team owner. Teams for Education allows administrators and teachers to create specific teams for classes, professional learning communities (PLCs), staff members, and everyone. Within a team, members can set up channels. Channels are conversation topics that allow team members to communicate without using email or SMS. Users can reply to posts with text as well as custom images, GIFs, and memes. Connectors are third-party services that can send information to the channel. Connectors include MailChimp, Facebook, Twitter, PowerBI and Bing News. Microsoft Teams allows teachers to share, provide feedback, and grade teamtaught student assignments using the Assignments area, available to Office 365 Education subscribers. Quizzes can also be assigned to students through an integration with Office Forms. Microsoft Teams relies on a number of Microsoft-specific protocols. Video conferences are carried out via the MNP24 protocol, known from the free versions of Skype. Skype for Business’s MS-SIP protocol is no longer used to connect Teams clients.

An Approach for Organising and Managing an Academic Year Using …

317

SIP and H.323-based VoIP and video conferencing clients require special gateways to connect to Microsoft Teams servers.

2.4 Google Meet Google Meet (formerly known as Hangouts Meet) is a video communication service developed by Google. It is one of two apps that replace Google Hangouts, the other being Google Chat. Google planned to begin retiring Google Hangouts in October 2019. In March 2020, Google rolled out Meet to personal (free) Google accounts. Free Meet calls can have only one host and up to 100 participants, compared to the 250 caller limit for Google Workspace users and the 25 participant limit for Hangouts. Unlike business calls with Meet, consumer calls are not recorded and stored, and Google says consumer data from Meet will not be used for ad targeting. While call data is not used for advertising purposes, based on a review of Meet’s privacy policy, Google reserves the right to collect data regarding the duration of the call, who is participating, and the IP addresses of the participants. Users need a Google Account to initiate calls, and like Google Workspace users, anyone with a Google Account can initiate a Meet call from within Gmail. From March 2021, meetings for free accounts will be limited to 60 min each.

2.5 Discord Discord is a digital distribution platform designed to create communities from gamers to education and businesses based on the type of instant messaging application that also incorporates VoIP service. Discord specializes in text, image, video and audio communication between users using chat channels and runs on Windows, MacOS, Android, iOS, Linux, and in web browsers. As of 2019, there are over 250 million users of the software. Discord is specially designed for use during gaming, as it includes features such as low latency, free voice chat servers for users, and dedicated server infrastructure. Video calling and screen sharing features were added to Discord, first for a small test base in August 2017 and later for all users in October 2017. While these features mimic the live streaming capabilities of platforms like Twitch [8], the company does not intend to compete with these services, believing that these features are best used by small groups (up to 25 users). The Discord platform has been implemented in: JavaScript, React, Elixir, Rust.

318

L.-A. Scutelnicu and M. C. Ceobanu

Table 1 Comparison between the five platformed descried in previous sub-sections Platform

Free version

No. meeting participants (default)

Screen sharing

Discord

Yes

25 (on video) Yes

White board

Meeting record

E2E encryption

Mobile app

No

No

No

Yes

500 (on audio) Zoom

Yes

100 (video and audio)

Yes

Yes

Yes

No

Yes

Microsoft Teams

Yes

250 (video and audio)

Yes

Yes

Yes

No

Yes

Google Meet

Yes

100 (video and audio)

Yes

No

Yes

No

Yes

Cisco Webex

Yes

200 (video and audio)

Yes

Yes

Yes

Yes (optional)

Yes

2.6 Comparisons Between the Mentioned Platforms Based on the short brief description of the text and video streaming platforms, we made a short comparison between them (see Table 1). As can be seen from Table 1, all of the platforms do approximately the same thing, but have certain limitations, especially on the number of meeting participants. But this issue, as we mentioned in the beginning of this chapter, can be removed with a paid subscription.

3 The Main System—Algorithmic Proposal Due to the many platforms that are available both free and paid, we decided that the platform that can fulfill our problems is based on Discord. The decision was made because, Discord is the only platform that allows external users and developers to integrate additional programs or code snippets or APIs, that can cover almost the use-cases that we took into consideration. Another aspect that attracted us to use this platform to took it into consideration, was that it is an open-source (free) platform and it has the possibility and offers the opportunity to implement so-called bots, which are in fact programs implemented by developers, which perform certain processes automatically or semi-automatically, with certain pre-established conditions at the time of development. Discord uses servers and channels similar to Internet Relay Chat (IRC) [9], even though these servers do not emulate traditional hardware architecture due to its distributed nature. A user can create a server on Discord, manage their public visibility and access, and create one or more channels within this service. Within a server, depending

An Approach for Organising and Managing an Academic Year Using …

319

on access controls, users can create channels within a framework of categories, with channel visibility and access customizable for the server as well. One such customization is the ability to mark channels NSFW (Not Safe for Work), which forces first-time registered participants to channels to confirm that they are over 18 years old and wish to view such content. In addition to normal text-based channels, Discord servers can create voice chat channels. The Discord client is built on the Electron framework [10], using web technologies, which allows it to be cross-platform and run on the web and as an app installed on personal computers. The software is supported by eleven data centers around the world to keep latency with clients low. Discord developers also added video calling and screen sharing since 2017. Support for calls between two or more users was added in an update in 2016. In December 2016, the company introduced the Game Bridge API, which allows game developers to integrate directly with Discord within games. Documentation for the Discord API is free and hosted on GitHub [11]. Discord provides partial support for rich text via Markdown syntax [12]. Discord uses the Opus audio format [13], which has low latency and is designed to compress the audio signal. In our study, we developed a bot (written in Go language [14]), that satisfies the needs of organising a discipline with specific user-roles for professors, students, groups etc. To create a bot, we needed to create an application for Discord. A bot is then attached to this application. Each bot, in turn, has a TOKEN attached, that Discord automatically recognizes when it is used in APIs. It must remain secret and, if it is made public, it must be reset as soon as possible because anyone who has access to the token can modify the bot. The bot can act on some members or channels only if it is active, therefore it must run permanently on a machine, a server. Also, there is a limit of 100 servers, if the limit is reached the bot owners must verify the bot by providing Discord with their ID so that Discord can control the bot and take legal action in case of TOS abuse [15]. To have a better control on the discipline server, we have used a database (developed in MongoDB [16]), which has the structure for the bot as follows: • special_roles with the following structure: {id: ‘’}. Any user whose discord id is entered in this collection will be able to actively interact with the bot: they can issue commands; • servers with the following structure: an example of the big server roles are attached to their Discord ids { _id: ‘891080993962856468’, // discord server id name: ‘FII Server 2021-2022’, // server name roles: {// roles along with their discord ids ‘1MSI’: ‘891080994038362127’, ‘1MIA’: ‘1026225507878060112’,

320

L.-A. Scutelnicu and M. C. Ceobanu MIA: ‘1026226278828871802’, ‘2B4’: ‘891080994088697889’, year1: ‘891080994109685776’, ‘1A5’: ‘891080993988018194’, ‘1A1’: ‘891080994009014314’ .... } }

• stud_ids_to_roles collection that holds information about student roles with the following structure: { _id: ‘’, roles: [‘verified’, license, ‘year2’, ‘e_series’, ‘2E1’, ‘2CDC4’], // roles nickname: ‘John Doe’, // the nickname set automatically discord_id: ‘’ // initially missing and filled in automatically }

All data are processed automatically using python and bash scripts. The data about the students are provided by the faculty secretariat and all the data are transformed into .json file format and then imported into the database using MongoDB Compass. In order for a user to have access to the server, he must enter it. Initially, he will only have access to the #unverified channel where he can see the instructions to receive roles. He sends a message to the bot that he sees in the list of active members with ##. If it exists in the stud_ids_to_roles collection then discord_id is filled with its discord id and the roles are taken from its role list and then joined with the server’s role list to get their ids and are offered to it. A discord_id is attached to a document and a person’s roles only if that discord_id coincides with the one that was set the first time when the bot’s registration number was sent in a private message or the discord_id is null. Depending on the roles, the student has access to certain channels and servers, for example: people with the a_ series role has access to the A series channels on any server that offers that role. If a person has the role 2B3, he has access only to the servers that have that role and only to the channels allowed by him (2nd Year, B series, 3rd Group). The bot also has a functionality through which servers are created automatically. The creation process is as follows: • to create a master or optional server, the command is: masopt, , , ,

• to create the license server, the command is:

An Approach for Organising and Managing an Academic Year Using …

321

license, , , , , , , Initially, we sent messages with the help of the bot to all the people who joined the servers where it was active. But this behavior is abusive and must be avoided because it is spam and can be considered a mass phishing attempt. After the bot was banned, we had to contact Discord and communicate that our bot does not commit crimes and that it needs to be unblocked, because it is used for educational purpose. They unlocked it and suggested that it is good to remove that message mechanism and the best option is that each user to send a private message to the bot for registration. A bot can have abusive behavior through: mass bans, mass kicks, mass messages, anything done in large quantities. Having these scripts developed, we created the Discord servers related to the study disciplines of the academic year related to the Faculty of Computer Science, which, in addition to the roles of teacher or student, also has an administrator role, this being owned by the course coordinator. This allows the administrator to make manual changes in addition to the ones we implemented automatically. The automation comes both by the fact that the students data are taken from certain lists and transposed into a database, but also by the fact that the Discord server related to the discipline receives a predefined number of groups by the owner or creator of the server (through one of the commands mentioned above), students receive the roles related to the groups they belong to and are automatically distributed to their interest group. This saves both the teachers and the system administrators a lot of time organizing the discipline.

4 Conclusions In the present work, we made an overview of the online streaming platforms, in which we went through various aspects related to their advantages and disadvantages of their usage. Most have free user support, but with certain limitations, related to the number of users who can be connected simultaneously in a video and or audio session, the quality of the image, the number of minutes allocated to a work session etc. These aspects can be eliminated through monthly or annual paid subscriptions. Due to the SarsCov2 pandemic, which intervened in 2020, we chose to transfer the classes of our Faculty of Computer Science “Al. I. Cuza” University of Ias, i, Romania, from the physical environment to the online one. After an analysis, we concluded that the best platform that fits our needs is Discord. It has support for the integration of so-called bots (programs implemented by developers), which allowed us to develop these programs for the organization of the academic year, with a certain limited access depending on the type of users who can connect to such a machine (so-called Discord server).

322

L.-A. Scutelnicu and M. C. Ceobanu

In the first year of application of this system, approximately 100 Discord servers for disciplines from bachelor and master were implemented, of which approximately 60 were intensively used. In the following year, the requirement increased, and servers for optional disciplines also appeared, thus, the total number of disciplines was approximately 140, of which 125 were intensively used. The pandemic has faded, the number of cases has drastically decreased both through vaccination and immunization, and we have returned to the physical education system, but the demand for this platform is still high. This shows the fact that communication between teachers and students is much easier and eases work times and sending and sharing information, comparable to the time allocated to an email conversation.

References 1. Munir, H., Vogel, B., Jacobsson, A.: Artificial intelligence and machine learning approaches in digital education: a systematic revision. Information, 13, 203 (2022). https://doi.org/10.3390/ info13040203 2. Chandankhede, A.: E-learning in education. Int. J. Adv. Res. Sci. Commun. Technol. 32–35 (2022). https://doi.org/10.48175/IJARSCT-7407 3. Holmes, C., Goldstein, A., Rasmussen, L., Robertson, L., Crits-Christoph, A., Wertheim, O., Anthony, J., Barclay, S., Boni, F., Doherty, C., Farrar, J., Geoghegan, L., Jiang, X., Leibowitz, L., Neil, J.D., Skern, T., Weiss, R., Worobey, M., Andersen, G., Garry, F., Rambaut, A.: The origins of SARS-CoV-2: a critical review 184(19), 4848–4856 (2021). ISSN 0092-8674. https:/ /doi.org/10.1016/j.cell.2021.08.017 4. Meyn, J., Kandziora, M., Albers, S., et al.: Consequences of platforms’ remuneration models for digital content: initial evidence and a research agenda for streaming services. J. Acad. Mark. Sci. 51, 114–131 (2023). https://doi.org/10.1007/s11747-022-00875-6 5. Schollmeier, R.: A definition of peer-to-peer networking for the classification of peer-to-peer architectures and applications. In: The 1st IEEE International Conference on Peer-to-Peer Computing (P2P), August 2001, pp. 101–102 (2001) 6. Javad, P., Sina, K., Edalatpanah, S.A.: A survey of voice over internet protocol (VOIP) technology. Int. Comput. Math. Sci. Appl. (2012) 7. Chaimeeboon, J., Namee, K.: Implementation a WebEx conferencing Testbed for DLIT classroom. In: Conference: The 9th International Conference on Sciences, Technology and Innovation for Sustainable Well-Being (STISWB 2017) at: Yunnan, China (2017) 8. Twitch platforms. https://www.twitch.tv/ 9. Emad, S., Zhen, J., Ahmed, E.H.: On the Use of Internet Relay Chat (IRC) Meetings by Developers of the GNOME GTK Plus Project, pp. 107–110 (2009) https://doi.org/10.1109/ MSR.2009.5069488 10. Electron Framework. https://www.electronjs.org/docs/latest/ 11. Discord API GitHub. https://github.com/topics/discord-bot-host 12. Markdown Syntax. https://daringfireball.net/projects/markdown/syntax 13. Opus Audio Format. https://opus-codec.org/h 14. Go Language. https://go.dev/ 15. Discord Terms and Conditions. https://discord.com/terms 16. MongoDB. https://www.mongodb.com

Exploring the Impact of COVID-19 on Education: A Study on Challenges and Opportunities in Online Learning Ananga Thapaliya and Yury Hrytsuk

Abstract The COVID-19 pandemic has brought about numerous changes in the education sector, including the transition to online learning. This literature review explored the impact of the pandemic on higher education and the challenges and opportunities associated with online learning. Results indicate that students and teachers in developing nations faced significant difficulties due to the lack of proper digital infrastructure. Additionally, a dearth of information was found regarding the impact of COVID-19 on academic performance. Nevertheless, the pandemic also brought about opportunities for innovation and capacity building in the education sector. The study concluded that neither teachers nor students were fully prepared for the shift to online learning during the pandemic. This literature review highlights the need for future research to investigate the effect of COVID-19 on academic performance and to develop strategies to improve online learning experiences for all students.

1 Introduction Globally speaking, the COVID-19 pandemic has had an influence on almost every nation and territory. In Wuhan, China, the outbreak’s initial cases were noted in December 2019 [1]. Governments from all around the world have urged the populace to practice social distancing, wash their hands frequently, wear masks, and steer clear of big gatherings as preventative measures. To contain and slow the spread of the illness, lockdowns, and stay-at-home directives have been put in place [2]. tHE PANDEMIC HAD A profound effect on all aspects of life, including the economy, and it did not just afflict underdeveloped nations [3, 4]. Although every country was impacted, it was projected that developing nations will be hit the worst since A. Thapaliya (B) Innopolis University, Innopolis, Russia e-mail: [email protected] A. Thapaliya · Y. Hrytsuk Schaffhausen Institute of Technology, Schaffhausen, Switzerland © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_29

323

324

A. Thapaliya and Y. Hrytsuk

they lack resources including infrastructure, technology, and healthcare facilities [5]. The inequality gap across nations, regions, and communities would enlarge as a result [5, 6]. The United States, China, Europe, Iran, South Korea, and other regions were among those first severely hit by significant pandemics. Authorities in several nations have put processes and rules into place to try and stop the spread of COVID-19. This includes putting in place measures like restricting social gatherings and encouraging social segregation through widespread lockdowns [7]. In order to combat the virus, a number of social and commercial activities, including gyms, museums, movie theaters, swimming pools, and educational institutions, had to be suspended [8]. The increasing expansion of COVID-19 has also provided a significant issue for the educational field, as educational institutions at all levels were compelled to close and discover new means of instructing and learning [9, 10]. The COVID-19 laws have forced educational institutions all around the world to adopt online learning because traditional classroom-based learning is no longer practicable [11, 12]. It is clear that COVID-19 has seriously disrupted the educational system, much of which is still being comprehended as a result of its extensive ramifications [13, 14]. The transition from in-person to online learning has many stakeholders, including government officials, academic staff, students, and parents, worried about the possible results [15]. Although the increased use of online learning brings with it new difficulties, the potential for innovation in the education field should not be disregarded. We seek to answer the following research questions from our study: 1. What are the challenges and opportunities associated with the transition to online learning due to COVID-19 in the higher education sector? 2. How has the COVID-19 pandemic affected the education sector in developing nations in terms of digital infrastructure and online learning capabilities? 3. What is the impact of the COVID-19 pandemic on academic performance in the higher education sector, and what evidence supports this impact? 4. What are the gaps in the current literature on the impact of COVID-19 on higher education and online learning, and what areas require further research and investigation? This study seeks to offer a comprehensive assessment of the available literature on the subject in light of the numerous concerns that have been voiced regarding the quality of online teaching and learning. During the shutdown and COVID-19 pandemic, it is important to understand the practices, difficulties, and opportunities related to online teaching and learning. It outlines the difficulties and possibilities of online learning and continuing education throughout the pandemic and offers suggestions for the future. This paper is organized as follows. Section 2 describes the methods we used for the literature review, Sect. 3 explains the results we obtained for the above research questions by grouping our findings into methods for the continuation of online education, obstacles for learning and teaching and opportunities for teaching and learning, Sect. 4 discusses the findings and Sect. 5 concludes the study and provides some insights for future research.

Exploring the Impact of COVID-19 on Education …

325

2 Methodology To locate pertinent studies, a thorough search was carried out across 10 databases (Google Scholar, Microsoft Academic, Education Resources Information Center, Research Gate, Semantic Scholar, and JSTOR) [16]. Additional pertinent studies were located using the reference lists of papers that were discovered during the original search. COVID-19, coronavirus, online learning, e-learning, and e-teaching were among the search phrases utilized. The most pertinent papers for review were chosen using inclusion and exclusion criteria that were created. The research was to deal with the potential and challenges presented by higher education’s use of online learning and teaching during COVID-19. 380 peer-reviewed publications were picked from the 500 articles gathered in the initial search, and 50 of them were chosen for the final review based on their abstracts. The publications had to meet a number of requirements, including publication on an accredited source, explicit research aims related to the current topic, and appropriate methodology. The final review only included full-text papers.

3 Results This conclusion was reached after reviewing the literature on online learning during the COVID-19 pandemic. Few of the studies that were reviewed emphasized the opportunities and academic outcomes that the pandemic presented, with the majority of them concentrating on how educational institutions responded to it and the challenges of online learning. The findings are divided into three groups: methods for continuing education online, obstacles for teaching and learning, and opportunities for teaching and learning.

3.1 Methods for the Continuation of Online Education Worldwide, educational establishments, including schools and higher education institutions, have been forced to close due to the COVID-19 pandemic and the ensuing lockdown and social isolation measures [17]. As a result, there has been a change in how educators deliver education, with an increase in the usage of different online platforms. Despite the difficulties that both teachers and students face, online learning, distance learning, and continuing education have emerged as the answers to this worldwide dilemma. Both students and teachers may find the switch from in-person to online learning to be novel and unfamiliar, but it is a need they must adapt to with few options available [11, 18]. In spite of their lack of readiness, the educational system and educators are currently implementing online platforms.

326

A. Thapaliya and Y. Hrytsuk

Despite the shutdown of educational facilities, online education technology has played a significant role in helping schools and universities continue to educate students during the pandemic [19, 20]. In order to make the transfer to this new method of learning successful, it is imperative to evaluate and support staff and student readiness. While some students with a fixed perspective can find it difficult to adjust, those with a growth mindset typically find it easier to adjust to different learning situations. There is no one-size-fits-all strategy for online learning because the needs of various disciplines and age groups vary [21]. Online education also reduces the requirement for physical mobility by giving students with physical disabilities more independence and the opportunity to participate in virtual learning activities [12, 22]. Globally, the COVID-19 pandemic has had a profound influence on education, resulting in the closure of several schools. This unforeseen circumstance has an impact on students, parents, and teachers. Education systems work to guarantee pupils continue to receive a high-quality education despite efforts by governments and health officials to stop the virus’s spread. However, many students find it challenging to concentrate on their academics when they are at home and under psychological and emotional stress [23]. The knowledge and expertise of both instructors and students with information and communications technology (ICT) may impact different approaches to online education. Many platforms have been used to deliver education, training, and skill development programs, including Google Classroom, Microsoft Teams, Canvas, and Blackboard [19]. These platforms make it simpler to organize classes and interact by providing tools like chat, video conferencing, and file storage. They facilitate the exchange of a variety of material formats, including Word documents, PDFs, Excel files, audio, and video, among others. Additionally, these platforms give users the option to analyze and monitor student learning through quizzes and the grading of submitted work using a predefined rubric [24]. A session of in-depth conversation and participation with instructors and classmates is followed by the distribution of learning materials, such as articles, videos, or links before the class even starts. The promotion of critical thinking, problemsolving, and self-directed learning is greatly enhanced by this approach. The usage of online learning management systems (LMSs) in the cloud, such as Elias, Moodle, BigBlueButton, and Skype, is becoming more common [25].

3.2 Obstacles for Learning and Teaching The COVID-19 pandemic has significantly disrupted the educational systems around the world, which has resulted in a wide acceptance of distance learning as a method of continuing education. While there are numerous benefits to remote learning, there are also new difficulties that have emerged, notably in terms of barriers to teaching and learning. Numerous challenges have arisen in education as a result of the abrupt change to online learning brought on by COVID-19 and lockdowns [26]. While some institutions that already had established online systems saw success, many

Exploring the Impact of COVID-19 on Education …

327

universities and colleges found it challenging to make the switch from face-to-face to online instruction [27]. The main difficulties that teachers and students had when using remote learning during the COVID-19 pandemic are explored in this research study. Lack of access to dependable technology and high-speed internet is one of the greatest problems that teachers and students in remote learning encounter [28]. In low-income households, where students do not have access to the gear and internet connectivity required for efficient remote learning, this problem is particularly acute. Due to the fact that students without access to technology are less likely to engage in remote learning and may lag behind their peers, the digital gap has major effects on student results [29]. In addition to being pleasurable, going to school helps kids develop their social skills and awareness. But altering the regular school timetable can have negative effects on kids’ life on financial, social, and psychological levels [30]. Students are more susceptible to online exploitation as more take online classes and spend more time on digital platforms. The risk of being exposed to potentially harmful or violent information, as well as cyberbullying, grows with this prolonged and unstructured use of online learning [31]. Families are depending more on technology and digital solutions to keep their kids interested in learning, entertained, and connected to the outside world as a result of school closures and strict containment measures. But not all kids have the knowledge and tools needed to stay safe online. The lack of proper digital infrastructure, particularly in schools and colleges, is a significant barrier to efficient remote learning [32]. Institutions frequently lack the resources or knowledge required to offer a reliable and secure digital platform for distance learning, which can cause technical issues and disruptions to the learning process. This can make students and teachers frustrated and may have a negative effect on student outcomes and mental health [33]. Lack of teacher training and online education delivery experience is another obstacle to efficient remote learning [34]. Many teachers could find it challenging to modify their teaching strategies for the online setting because they lack prior expertise with remote learning technologies [35]. Students may encounter low-quality remote learning as a result, of losing interest and motivation to engage in online learning. Due to the lack of academic outcome measurements at the time, there is little information available on the effect of the COVID-19 pandemic on student academic performance. However, it is anticipated that further research will be done on this subject as academic results become accessible. A study [36] examined student academic performance before and after COVID-19 imprisonment and discovered that students’ scores significantly improved, even on examinations that had previously been administered online. It is clear that the new assessment process was not responsible for the improvement because it was only noticeable when comparing data from after the confinement. Although this study did not identify a direct connection between confinement and academic performance, it is anticipated that the COVID-19 outbreak will increase dropout rates in addition to low student performance [37, 38].

328

A. Thapaliya and Y. Hrytsuk

3.3 Opportunities for Teaching and Learning The quick transition to online learning is just one example of the dramatic changes to the educational system brought forth by the COVID-19 pandemic. This has brought about a number of difficulties for the education sector, but it has also opened up chances for expansion and improvement. Universities have been forced to adopt novel strategies and technologies as a result of the abrupt change to online learning, which has sparked the creation of fresh industry solutions [39]. The difficulties presented by online instruction and learning have also encouraged innovative thinking and fresh approaches [11, 24]. Increased exposure to internet platforms will also allow students and teachers to improve their interpersonal and online communication abilities [40]. The COVID-19 pandemic has also created openings for brand-new study topics, such as the utilization of digital data collection techniques and improved accessibility to online research dissemination [41, 42]. Despite the considerable difficulties brought on by COVID-19, there are some chances for expansion and improvement in the education industry. While the transformation has caused many difficulties, it has also opened up new teaching and learning opportunities. We will look at some of these prospects in this literature study. Online learning offers more flexibility and accessibility, which is one of its key advantages. Students can engage in classes from anywhere with an internet connection thanks to online courses, which remove any time or geographic restrictions. Students who encounter obstacles to traditional in-person learning, such as those with disabilities, students in rural locations, or students with schedule issues, have benefited greatly from this [43]. Additionally, the flexibility of online learning allows students to study course materials at their own pace, which can enhance their comprehension and recall of the information [44]. Students can also customize their learning experiences to suit their unique requirements and preferences by using online learning, which also offers options for personalized learning. For instance, some online learning environments give students the option to set their own pace for coursework completion or to concentrate on particular areas of interest. This individualized method may increase pupils’ motivation and involvement [45]. Technology use in the classroom has expanded as a result of the switch to online education. This has made it possible to incorporate cutting-edge teaching strategies into the learning process, such as gamification, virtual reality, and simulation [46]. Students’ enthusiasm and engagement can increase when technology is used in the classroom, and their critical thinking and problem-solving abilities can also improve [47]. The limitations of virtual instruction must be overcome by teachers using creative strategies. To improve their online teaching methods, they are collaborating inside their neighborhood. As educators, parents, and children all go through comparable experiences, this offers unmatched potential for collaboration, creative solutions, and a readiness to learn from others and try out new methods [21]. To help create a more engaging and participatory learning environment, several educational institutions give up their resources and solutions.

Exploring the Impact of COVID-19 on Education …

329

4 Discussion The COVID-19 pandemic has had a significant effect on higher education, resulting in several disruptions and difficulties. It has, however, also offered some exceptional teaching and learning possibilities. For instance, the move to online teaching and learning has compelled institutions to investigate fresh ideas and enhance their digital expertise. Additionally, this has made it possible for teachers to work together and share expertise while also developing fresh and interesting online learning experiences [48]. Additionally, the pandemic has produced chances for digital capacity building and research. While there is no denying that virtual education presents obstacles, research indicates that the pandemic has also spurred the development of innovative and collaborative solutions as well as a willingness to adopt new tools and techniques [49]. In the end, the COVID-19 pandemic has sparked a transformation in the education industry, providing both difficulties and chances for instruction and learning [11]. This study clarifies the fact that the COVID-19 pandemic has created a number of difficulties for both students and teachers when it comes to online learning. According to this study, these difficulties are particularly severe for students in developing nations who might not have access to the necessary internet infrastructure. The study also discovers a dearth of information on the effect of COVID-19 on academic performance, which could be investigated in subsequent research as more information becomes available [24]. The research acknowledges that, despite the difficulties, the pandemic has also provided chances for innovation and capacity building. The study comes to the conclusion that during the pandemic, neither teachers nor students were sufficiently prepared for the shift to online learning [43, 46].

5 Conclusion and Future Research The problems and opportunities brought about by the pandemic and its effects on the higher education sector are highlighted in this literature review’s conclusion. The results indicate that students and academics, particularly in developing countries, who were not adequately prepared for the transition to online learning, have faced a number of interrelated issues as a result of the pandemic. Overall, this review of the literature emphasizes the necessity of continuous innovation and adaptation in the higher education sector to successfully address the issues and take advantage of the opportunities brought about by the COVID-19 pandemic. The current study makes it evident that there are many potential and obstacles related to online learning because of COVID-19. On the academic achievement of pupils throughout this time period and its effects, there is, however, scant study evidence. Future studies should therefore concentrate on examining how students fared in their academic endeavors throughout the pandemic and the elements that contributed to their success or failure. The efficiency of the various methods and

330

A. Thapaliya and Y. Hrytsuk

delivery mechanisms that have been employed to deliver online learning as well as the degree to which they have addressed the issues could also be explored in this study. Research could also examine the effects of online education on students’ mental health and well-being, particularly in underdeveloped nations where students are more likely to encounter several difficulties. When navigating the new reality of online learning during the pandemic and beyond, educators, policy-makers, and institutions may find the findings of such studies to be quite helpful.

References 1. Kumar, A., Singh, R., Kaur, J., Pandey, S., Sharma, V., Thakur, L., Sati, S., Mani, S., Asthana, S., Sharma, T.K., et al.: Wuhan to world: the covid-19 pandemic. Frontiers Cellular Infect. Microbiol. 242 (2021) 2. Sintema, E.J.: Effect of covid-19 on the performance of grade 12 students: implications for stem education. Eurasia J. Math. Sci. Technol. Educ. 16(7), em1851 (2020) 3. Chakraborty, I., Maity, P.: Covid-19 outbreak: migration, effects on society, global environment and prevention. Sci. Total Environ. 728, 138882 (2020) 4. Settersten, R.A., Jr., Bernardi, L., Härkönen, J., Antonucci, T.C., Dykstra, P.A., Heckhausen, J., Kuh, D., Mayer, K.U., Moen, P., Mortimer, J.T., et al.: Understanding the effects of covid-19 through a life course lens. Adv. Life Course Res. 45, 100360 (2020) 5. Blundell, R., Costa Dias, M., Joyce, R., Xu, X.: Covid-19 and inequalities. Fiscal Stud. 41(2), 291–319 (2020) 6. Jæger, M.M., Blaabæk, E.H.: Inequality in learning opportunities during covid-19: Evidence from library takeout. Res. Soc. Stratification Mobility 68, 100524 (2020) 7. Han, E., Tan, M.M.J., Turk, E., Sridhar, D., Leung, G.M., Shibuya, K., Asgari, N., Oh, J., García-Basteiro, A.L., Hanefeld, J., et al.: Lessons learnt from easing covid-19 restrictions: an analysis of countries and regions in Asia pacific and Europe. The Lancet 396(10261), 1525– 1534 (2020) 8. Andriani, H., et al.: Effectiveness of large-scale social restrictions (psbb) toward the new normal era during covid-19 outbreak: a mini policy review. J. Indonesian Health Policy Adm. 5(2) (2020) 9. Hebebci, M.T., Bertiz, Y., Alan, S.: Investigation of views of students and teachers on distance education practices during the coronavirus (covid-19) pandemic. Int. J. Technol. Educ. Sci. 4(4), 267–282 (2020) 10. Jeste, S., Hyde, C., Distefano, C., Halladay, A., Ray, S., Porath, M., Wilson, R., Thurm, A.: Changes in access to educational and healthcare services for individuals with intellectual and developmental disabilities during covid-19 restrictions. J. Intellect. Disability Res. 64(11), 825–833 (2020) 11. Adedoyin, O.B., Soykan, E.: Covid-19 pandemic and online learning: the challenges and opportunities. Interact. Learn. Environ. 1–13 (2020) 12. Paudel, P.: Online education: benefits, challenges and strategies during and after covid-19 in higher education. Int. J. Stud. Educ. 3(2), 70–85 (2021) 13. Bryson, J.R., Andres, L.: Covid-19 and rapid adoption and improvisation of online teaching: curating resources for extensive versus intensive online learning experiences. J. Geogr. High. Educ. 44(4), 608–623 (2020) 14. García-Morales, V.J., Garrido-Moreno, A., Martín-Rojas, R.: The transformation of higher education after the covid disruption: emerging challenges in an online learning scenario. Frontiers Psychol. 12, 616059 (2021) 15. Reimers, F.M.: Primary and Secondary Education During Covid-19: Disruptions to Educational Opportunity During a Pandemic. Springer Nature (2022)

Exploring the Impact of COVID-19 on Education …

331

16. Gasparyan, A.Y., Ayvazyan, L., Kitas, G.D.: Multidisciplinary bibliographic databases. J. Korean Med. Sci. 28(9), 1270–1275 (2013) 17. Chamberlain, L., Lacina, J., Bintz, W.P., Jimerson, J.B., Payne, K., Zingale, R.: Literacy in lockdown: learning and teaching during covid-19 school closures. Reading Teach. 74(3), 243– 253 (2020) 18. Henaku, E.A.: Covid-19 online learning experience of college students: The case of Ghana. Int. J. Multi. Sci. Adv. Technol. 1(2), 54–62 (2020) 19. Dash, S., Samadder, S., Srivastava, A., Meena, R., Ranjan, P.: Review of online teaching platforms in the current period of covid-19 pandemic. Indian J. Surg 84(Suppl 1), 12–17 (2022) 20. Subedi, S., Nayaju, S., Subedi, S., Shah, S.K., Shah, J.M.: Impact of e-learning during covid-19 pandemic among nursing students and teachers of Nepal. Int. J. Sci. Healthc. Res. 5(3), 68–76 (2020) 21. Doucet, A., Netolicky, D., Timmers, K., Tuscano, F.J.: Thinking about pedagogy in an unfolding pandemic: an independent report on approaches to distance learning during COVID19 school closures. Educ. Int. (2020) 22. Doyumgaç, I., Tanhan, A., Kiymaz, M.S.: Understanding the most important facilitators and barriers for online education during covid-19 through online photovoice methodology. Int. J. High. Educ. 10(1), 166–190 (2021) 23. Hasan, N., Bao, Y.: Impact of “e-learning crack-up”perception on psychological distress among college students during covid-19 pandemic: A mediating role of “fear of academic year loss”. Child. Youth Serv. Rev. 118, 105355 (2020) 24. Maatuk, A.M., Elberkawi, E.K., Aljawarneh, S., Rashaideh, H., Alharbi, H.: The covid-19 pandemic and e-learning: challenges and opportunities from the perspective of students and instructors. J. Comput. high. Educ. 34(1), 21–38 (2022) 25. Schneider, S.L., Council, M.L.: Distance learning in the era of covid-19. Arch. Dermatol. Res. 313(5), 389–390 (2021) 26. Toquero, C.M.: Challenges and opportunities for higher education amid the covid-19 pandemic: the philippine context. Pedagogical Res. 5(4) (2020) 27. Papapanou, M., Routsi, E., Tsamakis, K., Fotis, L., Marinos, G., Lidoriki, I., Karamanou, M., Papaioannou, T.G., Tsiptsios, D., Smyrnis, N., et al.: Medical education challenges and innovations during covid-19 pandemic. Postgrad. Med. J. 98(1159), 321–327 (2022) 28. Özüdo˘gru, G.: Problems faced in distance education during covid-19 pandemic. Participatory Educ. Res. 8(4), 321–333 (2021) 29. Hamid, R., Sentryo, I., Hasan, S.: Online learning and its problems in the covid-19 emergency period. J. Prima Edukasia 8(1), 86–95 (2020) 30. Tarkar, P.: Impact of covid-19 pandemic on education system. Int. J. Adv. Sci. Technol. 29(9), 3812–3814 (2020) 31. Shin, S.Y., Choi, Y.J.: Comparison of cyberbullying before and after the covid-19 pandemic in Korea. Int. J. Environ. Res. Publ. Health 18(19), 10085 (2021) 32. Tadesse, S., Muluye, W.: The impact of covid-19 pandemic on education system in developing countries: a review. Open J. Soc. Sci. 8(10), 159–170 (2020) 33. Zhai, Y., Du, X.: Addressing collegiate mental health amid covid-19 pandemic. Psychiatry Res. 288, 113003 (2020) 34. Chen, T., Peng, L., Jing, B., Wu, C., Yang, J., Cong, G.: The impact of the covid-19 pandemic on user experience with online education platforms in china. Sustainability 12(18), 7329 (2020) 35. Sutiah, S., Slamet, S., Shafqat, A., Supriyono, S.: Implementation of distance learning during the covid-19 pandemic in faculty of education and teacher training. Cypriot J. Educ. Sci. 15(1), 1204–1214 (2020) 36. Kuhfeld, M., Soland, J., Tarasawa, B., Johnson, A., Ruzek, E., Liu, J.: Projecting the potential impact of covid-19 school closures on academic achievement. Educational Res. 49(8), 549–565 (2020) 37. Dorn, E., Hancock, B., Sarakatsannis, J., Viruleg, E.: Covid-19 and student learning in the united states: the hurt could last a lifetime. McKinsey & Company 1, 1–9 (2020)

332

A. Thapaliya and Y. Hrytsuk

38. Moscoviz, L., Evans, D.K., et al.: Learning loss and student dropouts during the covid-19 pandemic: a review of the evidence two years after schools shut down. Cent. Glob. Dev. (2022) 39. Ratten, V.: Coronavirus (covid-19) and the entrepreneurship education community. J. Enterprising Communities: people Places Glob. Econ. 14(5), 753–764 (2020) 40. Daniel, S.J.: Education and the covid-19 pandemic. Prospects 49(1), 91–96 (2020) 41. Gillis, A., Krull, L.M.: covid-19 remote learning transition in spring 2020: class structures, student perceptions, and inequality in college courses. Teach. Sociol. 48(4), 283–299 (2020) 42. Russ, S., Hamidi, F.: Online learning accessibility during the covid-19 pandemic. In: Proceedings of the 18th International Web for All Conference, pp. 1–7 (2021) 43. Mseleku, Z.: A literature review of e-learning and e-teaching in the era of covid-19 pandemic (2020) 44. Huang, R., Liu, D., Tlili, A., Yang, J., Wang, H., et al.: Handbook on facilitating flexible learning during educational disruption: The chinese experience in maintaining undisrupted learning in covid-19 outbreak, vol. 46. Smart Learning Institute of Beijing Normal University, Beijing (2020) 45. Whalley, B., France, D., Park, J., Mauchline, A., Welsh, K.: Towards flexible personalized learning and the future educational system in the fourth industrial revolution in the wake of covid-19. High. Educ. Pedagogies 6(1), 79–99 (2021) 46. Pokhrel, S., Chhetri, R.: A literature review on impact of covid-19 pandemic on teaching and learning. High. Educ. Future 8(1), 133–141 (2021) 47. Ariawan, S., Malang, S.: Building critical thinking in covid-19 pandemic era: impossible or i am possible? Int. Res. J. Adv. Sci. Hub 2(6), 127–130 (2020) 48. Godber, K.A., Atkins, D.R.: Covid-19 impacts on teaching and learning: a collaborative autoethnography by two higher education lecturers. In: Frontiers in Education. Vol. 6, p. 647524. Frontiers Media SA (2021) 49. Zarzycka, E., Krasodomska, J.: Mazurczaka: Distance learning during the covid-19 pandemic: students’ communication and collaboration and the role of social media. Cogent Arts Humanit. 8(1), 1953228 (2021)

Memes as a Memorization Technique in Education Hamza Salem and Siham Siham Hattab

Abstract This paper explores the use of memes as a memorization technique in education. The study aims to define the concept of using memes as a memorization tool and to investigate its effectiveness in enhancing student learning outcomes. A review of literature is conducted to gather information on the current state of research in this area and to identify the key factors that contribute to the success of using memes as a memorization technique. The results of the study suggest that using memes as a memorization tool can be an effective way to promote student engagement and improve learning outcomes, particularly in subjects that are difficult to learn. Overall, this paper provides a comprehensive overview of the definition, benefits, and challenges of using memes as a memorization technique in education.

1 Introduction Memes have become a ubiquitous part of popular culture, widely shared and enjoyed on social media platforms, forums, and instant messaging apps. But what role can memes play in education. This paper explores the use of memes as a memorization technique in the classroom and their potential to enhance student learning outcomes. A meme is a cultural unit that is transmitted from one individual to another, typically through the internet. Memes are often humorous or satirical in nature and can take the form of images, videos, or texts. They often spread rapidly due to their shareability and can become a cultural phenomenon, with countless variations and adaptations appearing online [1]. There are a variety of memorization techniques that are commonly used in education. Some of the most common include repetition, mnemonics, visualization, and elaboration. Repetition involves repeating information multiple times in order to commit it to memory. Mnemonics involve creating a memorable phrase or acronym to help remember information. Visualization involves creating a H. Salem (B) · S. Siham Hattab Innopolis University, Innopolis, Russia e-mail: [email protected] S. Siham Hattab e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_30

333

334

H. Salem and S. Siham Hattab

mental image of information in order to remember it. Elaboration involves actively thinking about the information and connecting it to prior knowledge in order to reinforce memory. In comparison to these traditional memorization techniques, using memes as a memorization tool is a relatively new concept. The unique combination of humor and visual representation in memes has the potential to make information more memorable and engaging for students. This paper aims to provide a comprehensive overview of the definition and benefits of using memes as a memorization technique in education [2]. The goal of this study is to define the concept of using memes as a memorization tool, to review current research on the topic, and to examine the pedagogical and ethical considerations involved in using this technique in the classroom.

2 Literature Review The use of memes as a memorization technique in education is a relatively new concept and as such, there is limited research in this area. However, a review of the existing literature suggests that the use of humor and visual aids in memorization can be effective in enhancing student engagement and learning outcomes. One study found that the use of humor in educational settings can have a positive impact on students’ motivation and can increase their engagement in the material being taught [3]. The use of humor can also serve as a stress reliever, which can improve students’ overall well-being and ability to focus [4]. Additionally, research has shown that visual aids, such as images and videos, can be effective in enhancing memory retention and recall [5]. The use of visual aids in education has been shown to increase students’ ability to retain information, particularly in subjects that are difficult to learn [6]. The unique combination of humor and visual representation in memes makes them well suited for use as a memorization tool. A study [7] found that the use of memes in the classroom can increase student engagement and improve learning outcomes, particularly in subjects that are traditionally considered dry or boring. The study also found that students who used memes as a memorization tool reported higher levels of enjoyment and motivation in their coursework. In conclusion, the existing literature suggests that the use of humor and visual aids in education can be effective in enhancing student engagement and learning outcomes. The use of memes as a memorization tool has the potential to provide these benefits and more, given the unique combination of humor and visual representation that is inherent in memes. Further research is needed to fully understand the potential of memes as a memorization technique in education and to determine the best practices for their use in the classroom.

3 Methodology This study employed a pre-experimental design to explore the effectiveness of memes as a memorization technique in education. The pre-experimental design was chosen

Memes as a Memorization Technique in Education

335

because the goal of this study was to provide preliminary evidence of the effectiveness of memes as a memorization technique, rather than to establish cause-and-effect relationships. The study recruited a convenience sample of 80 students from a Russian university. All participants were between the ages of 18 and 25 and were enrolled in a Database course. The materials used in the study included 4 questions related to the content of the database course. These questions were used to assess the participants’ prior knowledge of the course content before the intervention and their recall after the intervention. In addition, we have created 4 memes shown in Figs. 1, 2, 3 and 4 related to the course content and used them as the memorization technique in the study. The data have been collected in face to face questions to check the student memorization for each term. The data were analyzed using descriptive statistics to compare the mean scores of the participants who were presented with the concepts in traditional format with those who were presented with the concepts in meme format.

4 Results The study aimed to examine the effectiveness of using memes as a memorization technique in education with a focus on database concepts. The results of the study are presented in Table 1, which shows the percentage of participants who correctly answered multiple-choice questions on various database concepts with and without the use of memes. The results show that the use of memes as a memorization tool significantly increased the participants’ understanding and recall of the database concepts. The participants’ average percentage of correct answers was higher with the use of memes compared to without the use of memes for all the database concepts

Fig. 1 Normalization

336

Fig. 2 Crows foot notations

Fig. 3 Normalization types Fig. 4 Functional dependency

H. Salem and S. Siham Hattab

Memes as a Memorization Technique in Education Table 1 Table of results Definition Normalization Crows foot notations Normalization types Functional dependency

337

With meme (%)

Without meme (%)

95 95 85 80

40 35 20 15

tested. For example, the participants had a 95% success rate in answering questions on normalization when using memes, compared to a 40% success rate without the use of memes. Similarly, the participants had an 85% success rate in answering questions on normalization types with memes, compared to a 20% success rate without the use of memes.

5 Conclusion The study aimed to investigate the effectiveness of using memes as a memorization technique in education, with a focus on database concepts. The results showed that the use of memes as a memorization tool significantly increased the participants’ understanding and recall of the database concepts. The participants’ average percentage of correct answers was higher with the use of memes compared to without the use of memes for all the database concepts tested. These findings suggest that using memes as a memorization tool can be an effective way to promote student engagement and improve learning outcomes in the field of database concepts. The use of humor and visual aids can help make the learning process more enjoyable and increase students’ retention of course content. However, the study had a small sample size and used a pre-experimental design, which limits the generalizability of the results. Further research with larger sample sizes and experimental designs is needed to provide more robust evidence on the effectiveness of memes as a memorization technique in education. Additionally, further studies could explore the potential challenges and limitations of using memes as a memorization tool. Overall, the results of this study provide valuable insights into the potential of memes as a memorization technique in education. The findings highlight the importance of incorporating humor and visual aids in the learning process and suggest that using memes as a memorization tool could be a promising approach for enhancing student learning outcomes in database concepts and other subjects.

338

H. Salem and S. Siham Hattab

References 1. Patrick, D.: The language of internet memes. Soc. Media Reader 120–134 (2012) 2. Egan, Kieran: Memory, imagination, and learning: connected by the story. Phi Delta Kappan 70(6), 455–459 (1989) 3. Salmee, S.A., Arif, M.M.: A study on the use of humour in motivating students to learn English. Asian J. Univ. Educ. 15(3), 257–265 (2019) 4. Purnama, Agnes Dian: Incorporating memes and instagram to enhance students participation. LLT J. J. Lang. Lang. Teach. 20(1), 1–14 (2017) 5. Paola, A.-A., et al.: Children’s memory of a story experienced with virtual reality versus traditional media technology 6. Brame, C.J.: Effective educational videos 1–5 (2015) 7. Kayali, N.K., Altuntas, A.: Using memes in the language classroom. Shanlax Int. J. Educ. 9(3), 155–160 (2021)

Quantifying Education in the Post-COVID Era: An Engineering Approach Concept Gerald B. Imbugwa and Tom Gilb

Abstract The outbreak of Covid-19 has had a catastrophic impact on the education sector, leading to an unprecedented number of students struggling to keep up with their curriculum. This paper proposes an engineering approach concept that incorporates stakeholder analysis, quantification, and planning to ensure that students receive a high-quality education. Through stakeholder analysis, educators can gain a better understanding of the needs and perspectives of the stakeholders and involve them in the development of effective educational strategies. The quantification component involves using data and metrics to objectively measure and evaluate educational outcomes, thereby supporting student learning and tracking progress. Planning, which is a critical component of the engineering approach, helps ensure that education remains resilient and adaptable.

1 Background 1.1 Problem The COVID-19 pandemic has had a profound impact on the traditional education system, forcing schools to close and sending students of all ages home. This has resulted in a shift to online learning platforms, which have presented multiple challenges to both students and educators. Students have found it difficult to adapt to the new platform, with its lack of direct instructional guidance hindering their ability to fully engage in the learning process. Teachers and parents have also had to grapple with the unpredictable quality and level of engagement, as well as the lack of ways G. B. Imbugwa (B) Innopolis University, Innopolis, Norway e-mail: [email protected] T. Gilb Independent Researcher, Kolbotn, Norway

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_31

339

340

G. B. T. Imbugwa Gilb

to measure student output due to the inability to directly observe progress. Furthermore, the lack of physical proximity has impeded meaningful conversations, robbing students of the opportunity for stimulating discourse.

1.2 Solution The use of an engineering approach concept, which would incorporate stakeholder, quantification, and planning. Stakeholder analysis involves identifying and engaging key groups who have a vested interest in the success of education, including parents, educators, administrators, policymakers, and community organizations [1]. By incorporating this aspect of the engineering approach, educators can better understand the needs and perspectives of the stakeholders and involve them in the development of effective education strategies [2]. Quantification, another key component of the engineering approach concept, involves the use of data and metrics to objectively measure and evaluate educational outcomes [3–5]. This helps educators track student progress, identify areas of weakness, and implement targeted interventions to support student learning [3]. With the use of technology and data, educators can ensure that students receive the education they need to succeed, even in a remote learning environment and post-COVID era [6]. Better planning is also an important part of the engineering approach concept [7, 8]. By developing detailed and flexible plans, educators can quickly pivot and adjust their approaches to meet the changing needs of students and communities [5]. This helps ensure that education remains resilient and adaptive in the face of change [8]. Considering the points above, a set of principles for the engineering approach concept can be established. This concept can guarantee high-quality education for students when confronted with difficulties [4, 5], through the understanding of stakeholders, establishment of quantified goals, and planning to fulfill these goals. This paper will be structured as follows: First, a literature review will be presented, followed by an application of the engineering approach concept through an example [9]. Then, a discussion of the methodology that will be employed to test this concept will be provided. Lastly, the conclusion will provide a general overview and plans for the next steps.

2 Literature Review The conventional approaches to teaching and learning have been severely disrupted by the COVID-19 pandemic. To maintain educational continuity, education systems around the world have been compelled to swiftly modify their approaches to teaching and learning [10]. Even though the pandemic has been difficult for both students and

Quantifying Education in the Post-COVID Era: An Engineering Approach Concept

341

teachers, research indicates that new methods have been developed. This article will go over some of the ways that the pandemic is altering how educational institutions teach, support learning, and assess academic achievement [11]. The increased use of technology to support learning in the classroom and at home is one benefit of the pandemic. For instance, online meetings, streaming, and virtual classrooms have become popular tools for teaching and parent-teacher communication in some nations [12]. With the aid of technology, students can interact with instructors and get feedback while relaxing in their own homes. In comparison to the conventional method of in-class instruction, this change has also made it challenging for teachers to monitor students’ progress [13]. The pandemic has promoted the creation of online learning platforms and virtual classrooms, among other remote learning technologies. The ability to access course materials, video lectures, and exam results from home is made possible by virtual classrooms. Additionally, online learning options for students all over the world have significantly increased thanks to virtual learning platforms like Khan Academy and Coursera [14]. The pandemic has compelled educators to consider novel ways to measure and evaluate learning. For instance, teachers are using online quizzes and rubrics more and more to keep track of their student’s progress. To evaluate student learning, schools also had to create alternative assessment techniques. For instance, some schools use project-based assessment to gauge how well students comprehend a subject or idea [13, 15]. In conclusion, the COVID-19 pandemic has forced academic institutions to fundamentally change how they influence knowledge and foster learning. By experimenting with new methods and applying an engineering mindset, the education system during this time can be enhanced even further. Our proposed concept can assist in ensuring that all stakeholders’ needs are met, in setting measurements for each goal or requirement, and in analyzing the results. More information on the idea and applications is provided in the following section.

3 Engineering Approach Concept The engineering approach is a systematic methodology for problem-solving that prioritizes rigour, well-defined processes, and quantification. The approach focuses on extreme clarity and heuristic-driven decision-making informed by a comprehensive systems overview. The engineering approach enables the identification, analysis, and solution of complex problems through objective data and evidence-based analysis. In contrast to subjective opinions or ambiguous assessments. The approach employs a structured and systematic approach to arrive at optimized solutions that are reliable, safe, meet the needs and expectations of stakeholders. The engineering approach is a disciplined and methodical problem-solving approach. Quantification and subsequent measurement are critical initial stages in achieving practical objectives. Education institution may suffer as a result of a lack of clear

342

G. B. T. Imbugwa Gilb

targets to strive for and precise criteria for evaluating success or failure. As a result, quantifying objectives is an important tool for motivating all the stakeholders in the academia in optimizing their efforts to achieve desired results [4]. Scale can be defined as central to the definition of all scalar attributes, that is, to all the performance and resource attributes [4]. A scale template for quantification is provided below. This scale can be used to assess and measure a variety of factors. The proper quantification of these factors can be critical to achieving set goals. The template allows for accurate progress measurement and the identification of areas that may require additional attention. Individuals and organizations can develop a comprehensive understanding of their performance and make informed decisions to improve their operations by implementing a clear and precise scale. In this sense, the use of a scale for quantification is an essential component of effective management and decision-making [4]. Tag: < Assign a tag to this Scale >. Type: Scale. Version: < Date of the latest version or change >. Owner: < Role/email of person responsible for updates/changes >. Status: < Draft, SQC Exited, Approved >. Scale: < Specify the Scale with defined [qualifiers] >. Alternative Scales: < Reference by tag or define other Scales of interest as alternatives and supplements >. In this example [9], a series of questions are posed to assess their difficulty and the effectiveness of the proposed approach in addressing them. A thorough understanding of the challenges faced and the identification of potential solutions can be facilitated by a systematic evaluation of the proposed approach in the context of the questions posed. As a result, the proposed method can be refined and optimized to achieve the desired result. The careful consideration of such questions is an essential component of effective problem-solving, and the proposed approach is intended to aid in this process. “It is important to use evidence for their appropriate creation and development so that action has the greatest impact possible.” From the excerpt, we can deduce the following questions: 1. 2. 3. 4. 5. 6. 7.

What is the significance of utilizing evidence in the context of the excerpt? What type of evidence is being referred to? What is the required quality of the evidence? Can you define the term ‘appropriate creation and appropriate development⣙? What specific actions are being referred to? How do we assess the impact of these actions? Are there any limitations, such as costs or long-term harm to students, that may limit the attainment of the greatest possible impact?

Quantifying Education in the Post-COVID Era: An Engineering Approach Concept

343

The absence of definitive responses and adequate clarity on the issues raised has become apparent. As a result, an engineering approach is required to effectively address these questions. This method is distinguished by a methodical and rigorous problem-solving methodology that emphasizes the use of quantifiable measures and well-defined processes. Below is a sample of how it could be done. Student Learning Capability: Ambition: “It is important to use evidence for their appropriate creation and development so that action has the greatest impact possible.” Scale: % improvement using [Learning Method] which has [Method Evidence] for [Learner Category] in [Learning Environment] in [Location]. Past: 0%, July 2020, Learning Method = Rote Memorization, Method Evidence = Mayer [16], Learner Category = Students, Learning Environment = Recognizing, Learning Environment = MIT Lab. Goal: 50%, July 2021, Learning Method = Rote Memorization, Method Evidence = Mayer [16], Learner Category = Students, Learning Environment = Recognizing, Learning Environment = MIT Lab. The proposed solution to the problems at hand involves the systematic application of an engineering approach. This method is intended to promote a rigorous and structured problem-solving methodology that places an emphasis on quantification and well-defined processes. This approach prioritizes a comprehensive overview of complete systems over a narrow technical focus, promoting extreme clarity through the use of heuristic-driven decision-making. In this case, the engineering approach aided in the quantification of the “impact” factor at 50% of the goal and the identification of at least six parameters, including the “learning method.” The original goals have also been transformed into five specific specifications, including tag, ambition, scale, past, and goal, to improve clarity and enable quantification during the goal formulation process. This methodical approach can be effective in the quantification and measurement of success in academia in the post-COVID era.

4 Methodology This research aims to examine the efficacy of the engineering approach in tackling the challenges confronting education in the post-COVID-19 period. A qualitative research methodology will be utilized as it offers a comprehensive exploration of the experiences, viewpoints, and beliefs of the participants. A purposive sample of 20 participants, including teachers, students, and administrators from the education sector, will be recruited to ensure diversity in backgrounds, experiences, and perspectives. Data will be collected through semi-structured interviews conducted online via video conferencing tools, owing to the COVID-19 pandemic. The interview questions will be devised based on the research objectives and will concentrate on the effectiveness of the engineering approach in addressing the challenges confronted by education in the post-COVID-19 era. Furthermore, data on

344

G. B. T. Imbugwa Gilb

educational outcomes from Innopolis University will be gathered and subjected to regression analysis to ascertain the correlation between the application of engineering approach concepts, metrics, and the improvement of educational outcomes. The obtained data will be transcribed verbatim and analyzed using thematic analysis, a commonly used technique in qualitative research. Thematic analysis involves the identification and examination of patterns or themes in the data through a continuous process of comparison and contrast. Several strategies, including member checking, peer debriefing, and triangulation, will be employed to guarantee the credibility of the research. Member checking involves sharing preliminary findings with the participants to ensure that their viewpoints are represented accurately. Peer debriefing involves sharing the outcomes with a colleague who possesses expertise in qualitative research to ensure that the analysis is thorough and impartial. Triangulation entails utilizing multiple sources of data, such as interviews, observations, and documents, to enhance the accuracy and reliability of the findings. In conclusion, this study will use a qualitative research approach to explore the effectiveness of the engineering approach in addressing the challenges faced by education in the post-COVID-19 era. Data will be collected through semi-structured interviews with a purposive sample of participants and analyzed using thematic analysis. The study’s reliability will be ensured by implementing various methods.

5 Conclusions The engineering approach is a rigorous and structured methodology for solving problems that prioritizes quantification, clearly defined procedures, and a comprehensive system overview. Adopting this approach can lead to the design of dependable, secure, and efficient educational solutions that meet the needs and expectations of all stakeholders. The post-COVID-19 era, which has significantly transformed the educational landscape and disrupted conventional teaching and learning methods, underscores the importance of this approach as a means to identify problems earlier and effectively address them. One key benefit of the engineering approach is its ability to promote data-driven decision-making. By quantifying goals and tracking progress, individuals can make informed decisions that are supported by evidence, rather than relying on arbitrary judgments or ambiguous assessments. This, in turn, leads to more effective and efficient educational settings that benefit both students and teachers equally. In addition, the engineering approach allows for the identification of areas for optimization and improvement. By methodically analyzing and enhancing teaching and learning approaches, more effective and productive educational programs can be designed that cater to the needs of both students and teachers. In conclusion, the engineering approach is a valuable tool for addressing the challenges facing education in the post-COVID-19 era. By promoting data-driven decision-making and systematic evaluation, optimized, risk-free, and efficient solu-

Quantifying Education in the Post-COVID Era: An Engineering Approach Concept

345

tions can be developed. Policymakers and educational institutions are advised to consider the use of this approach to address the challenges of education. The next steps involve testing the engineering approach concept and conducting interviews to determine the importance of this approach in educational settings.

References 1. Schiffer, J.R., Hoonakker, P.L.T.: Stakeholder engagement: a review of the literature. J. Bus. Res. 85, 59–68 (2018) 2. Lei, D., Skitmore, M.: Stakeholder Analysis in Construction Projects. Routledge (2018) 3. Black, Paul, Wiliam, Dylan: Classroom assessment and pedagogy. Assess. Educ. Principles Policy Pract. 25(6), 551–575 (2018) 4. Gilb, T.: Competitive Engineering: A Handbook for Systems Engineering, Requirements Engineering, and Software Engineering Using Planguage. Elsevier (2005) 5. Ming Tang, Y., Chung Chen, P., Law, K.M.Y., Wu, C.-H., Lau, Y.-Y., Guan, J., He, D., Ho, G.T.S.: Comparative analysis of student’s live online learning readiness during the coronavirus (covid-19) pandemic in the higher education sector. Comput. Educ. 168, 104211 (2021) 6. Chen, W., Liu, X., Yang, Y.: The use of technology in remote education: opportunities and challenges. J. Educ. Technol. Dev. Exch. 3(1), 1–14 (2020) 7. Gilb, T., Finzi, S., et al.: Principles of Software Engineering Management, vol. 11. Addisonwesley Reading, MA (1988) 8. Kerzner, H., Kerzner, P.: Project Management: A Systems Approach to Planning, Scheduling, and Controlling. Wiley (2017) 9. Themis CHRISTOPHIDOU.: European ideas for better learning: the governance of school education systems (2020) 10. Pokhrel, Sumitra, Chhetri, Roshan: A literature review on impact of covid-19 pandemic on teaching and learning. High. Educ. Future 8(1), 133–141 (2021) 11. Guo, P., Saab, N., Post, L.S., Admiraal, W.: A review of project-based learning in higher education: student outcomes and measures. Int. J. Educ. Res. 102, 101586 (2020) 12. Favale, T., Soro, F., Trevisan, M., Drago, I., Mellia, M.: Campus traffic and e-learning during covid-19 pandemic. Comput. Netwo. 176, 107290 (2020) 13. Arinal Rahman, M., Novitasari, D., Handrianto, C., Rasool, S.: Challenges in online learning assessment during the covid-19 pandemic. Kolokium J. Pendidikan Luar Sekolah 10(1), 15–25 (2022) 14. Rukmini, C.: Covid-19 pandemic–a paradigm shift in virtual learning. In: Covid-19 Pandemic: Issues on Sustainable Development Goals, p. 273 15. Kokotsaki, Dimitra, Menzies, Victoria, Wiggins, Andy: Project-based learning: a review of the literature. Improving Sch. 19(3), 267–277 (2016) 16. Bill & Melinda Gates Foundation. K-12 education (2023)

Teaching Object-Oriented Requirements Techniques: An Experiment Maria Naumcheva

Abstract Scenario-based software requirements specifications, due to limitations of natural language and scenarios, lack precision and abstraction. Formal methods address this problem, but are rarely used. A Unified Object-Oriented (OO) approach complements the simplicity and appeal of scenario techniques with the rigor and clarity of software contracts. In this study we conduct a teaching experiment to evaluate the perception of usefulness and difficulty of the approach. The obtained results demonstrate that the unified OO requirements approach has a potential to be adopted by requirements engineering practitioners.

1 Introduction In current practice, software requirements specifications heavily rely on natural language, scenarios [4, 9, 10] and UML diagrams [6]. Such specifications, due to limitations of natural language and scenarios, lack precision and abstraction. Formal methods address these limitations, yet their use is very limited outside of missioncritical systems [1, 6]. A Unified Object-Oriented (OO) approach [17] unites both worlds: it complements the simplicity and appeal of scenario techniques with the rigor of software contracts. In this approach, requirements, elicited as scenarios, are further analyzed to formulate abstract properties related to a system and its environment. The elements of a system and its environment are modeled as software classes, and requirements are formalized as software contracts of the features of those classes. The practical applicability of an approach depends on the required efforts for its learning. In this study we evaluate whether our proposed approach could be fitted into current university curriculum as a part of “OO Analysis and Design” course at the University of Toulouse.

M. Naumcheva (B) Innopolis University, Innopolis, Russia e-mail: [email protected] University of Toulouse/IRIT, Toulouse, France © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_32

347

348

M. Naumcheva

Modifying university curriculum is a sensitive topic. Since Covid19 outbreak, university programs experience tightened competition for students with massive open online courses (MOOCs) and online degree programs [16]. Students are focused on obtaining knowledge and skills that are relevant for their future jobs [11, 16]. As we aim to teach a new approach, which is not used in the industry yet, we should evaluate whether its learning helps to produce better specifications within commonly used approaches, such as use-case driven UML specifications. To address the concerns, discussed above, we formulate the following research questions for our study: (i) is limited training sufficient for learning OO contractbased requirements? (ii) does learning contract-based OO requirements techniques help to produce better UML specifications? The article is structured as follows. Section 2 describes the contract-based OO approach to requirements. Section 3 describes study design and results. In Sect. 4 we discuss related work, limitations and future work directions. Finally, we summarize the outcomes in Sect. 5 and conclude the paper.

2 Contract-Based OO Requirements In the unified OO approach to requirements, scenarios serve as an elicitation tool. In addition to commonly practiced requirements analysis techniques, such as UML diagrams [5], it involves formulating requirements as software contracts (preconditions, postconditions and class invariants). The contracts are formulated in a programming language which leads to seamless software development. The Eiffel language is a natural fit due to its readability and support of Design by Contract, yet the approach is applicable to any other OO language, such as Java or C++.

2.1 OO Requirements Fundamentals The basis of OO requirements is the concept of class, that serves as a unit of abstraction. The elements of the system and its environment are modeled as classes populated with relevant features (queries and operations). Those features are abstract (not implemented), as the classes describe system’s requirements, not implementation. The semantic specification of classes and their features is formulated with assertions, known as contracts [14]: preconditions express the properties that must hold when calling the routine, postconditions express the properties that must hold on the routine’s exit, class invariant expresses the properties that must be preserved by all methods of a class. [15] describes OO requirements fundamentals in details.

Teaching Object-Oriented Requirements Techniques …

349

2.2 Scenario-Driven OO Requirements The approach of Scenario-Driven OO Requirements relies on scenarios, since this technique is widely used in the industry [6]. After scenarios are identified, they are analyzed to extract abstract properties, such as environmental constraints, timeordering constraints, or specification of single operations. The scenarios themselves are modeled as specification drivers in a dedicated class. Each of scenario classes (use-case classes, test classes) includes a group of logically related scenarios, exercising features of other requirements classes. The detailed presentation of the approach with examples can be found in [17].

3 Teaching Experiment 3.1 Design The study is based on a controlled experiment [18] conducted as a part of the “OO Analysis and Design” course in the University of Toulouse, followed by a questionnaire. This course, as it is common to many universities [7], introduces UML as requirements modeling language. In total, 31 students participated in the experiment: bachelor’s students in their third year and master’s students in their first year. Course instructors provided a textual description of a case study to students, that they further used to elicit requirements and produce various requirements artifacts. The experiment was split into two parts. In the first part (1.5 h) theory on contractbased requirements was presented to students. They were already familiar with UML and scenario modeling. Further they had a task to describe two scenarios for each of the two given use cases according to a provided template. In the second part (4.5 h) students were randomly split into two groups and had to complete two tasks: 1. (2–3 h) Students of Group 1 specified contract-based OO requirements for the first use case. Students of Group 2 produced a sequence diagram for the first use case. 2. (2–3 h) Students of Group 1 worked on a sequence diagram for the second use case. Students of Group 2 specified contract-based OO requirements for the second use case. After submitting the results of their work, students filled in an online questionnaire. The questionnaire included two types of questions. Single-choice questions were formulated as statements that participants had to evaluate based on a Likert scale (‘Strongly disagree’, ‘Disagree’, ‘Agree’, ‘Strongly agree’, ‘No opinion’ choices). We used those questions to collect quantitative feedback on the use of the approach of Scenario-Driven OO Requirements. We also used open questions to collect additional qualitative feedback, such as (i) difficulties that the participants faced applying the approach and (ii) how suggested approach helped students to improve their UMLbased specifications.

350

M. Naumcheva

Fig. 1 Summary of evaluations of the statement “It was hard for me to formulate requirements as software contracts” (%, number of respondents)

3.2 Results All of the 31 study participants have participated in a survey, yet one of them answered only part of the questions. Although the population size is not large enough to draw definite conclusions, we have obtained a preliminary outlook on the 26% of respondents declared that it was hard to understand and use software contracts for requirements specification, whereas 43% had a positive experience and 30% were neutral. The particular difficulties, stated by the experiment participants, were: not enough familiarity with contracts; not enough examples provided; not enough practice (Fig. 1). All participants were able to list the advantages of formulating contract-based OO requirements. The questionnaire responses indicate that contract-based OO requirements techniques helped students to improve their UML specifications in the following ways: “to think of elements we hadn’t thought of, for example additional preconditions”, “to discover details that need to be added to features”, “to better define and implement use cases”, “to identify alternative scenarios for the system’s use cases”, “to better analyze the requirements in a global way for a specification”.

4 Discussion 4.1 Related Work Similarly to our approach, in the Unified Process (UP) [12] system operation contracts can be formulated to specify system operations, identified in use cases. System

Teaching Object-Oriented Requirements Techniques …

351

operation contracts describe how system’s operations change the state of the domain model objects and can be expressed in natural language or OCL. The UP does not discuss explicitly contracts describing environmental or time-ordering constraints. This approach does not generalize the notion of a class to model not only elements of a system and its domain, but also requirements artifacts, such as scenarios or tests. Briand et al. presents the results of two empirical studies on OCL contracts application. The first study [2] describes two controlled experiments with 38 participants. The second study [3, 13] describes a series of four controlled experiments with 33, 68, 21, and 61 participants respectively. Participants of both studies are 4th year bachelor’s students at Carleton University, who previously had extensive training in UML-based software development and are familiar with concepts such as preconditions, postconditions, and class invariants. Each of the studies took place over 4 labs which took about 5.5 h in total. The results of the studies are evaluated by comparing the correctness of obtained artifacts. The first study demonstrates that substantial, thorough training is required to obtain significant benefits by using OCL. In the second study statistically significant benefits of using OCL are obtained only in the fourth experiment, which the authors of the study connect with better training and motivation of the study subjects. Galinier [8] presents a case study with industry practitioners conducted to evaluate the RSML—a tool-supported domain-specific requirements language that produces contract-based specifications in Eiffel from the ones formulated in restricted natural language. Only 11 participants took part in a study which is significantly less than in studies with students of software engineering programs. Study participants specified requirements for a case study in RSML and filled in a questionnaire to evaluate their experience. Based on the questionnaire responses, the author concludes that the participants were able to learn how to use RSML fairly quickly.

4.2 Limitations The experiment is limited to a single course at the University of Toulouse, so the results might not be applicable to other educational settings or populations. We aim to conduct further studies to validate and generalize the obtained results. The difficulties reported by the participants, such as not enough familiarity with the Eiffel language and contracts, indicate that the course prerequisites might not have been explicit enough. This highlights the need for repeating the experiment with the improved supporting material, including more examples and illustrations, to help students better understand the concepts.

352

M. Naumcheva

4.3 Future Work The experiment, although limited in scope, provides valuable inputs for future empirical studies of the unified OO approach to requirements, such as the difficulties that study participants faced when applying the approach, listed in Sect. 3.2. We plan to conduct another experiment to evaluate whether these difficulties can be eliminated with improving the supporting material by including more examples and illustrations, and making explicit the course prerequisites (such as Design by Contract knowledge).

5 Conclusion Producing unambiguous requirements is an important problem in software engineering. To be embraced by the industry practitioners, the solution should satisfy two criteria: (i) it should not require substantial training, (ii) it should be perceived as useful for current practices. In the experiment, presented in this paper, we evaluated the unified OO approach to requirements against these two criteria. With respect to the study participants, both of the criteria were satisfied: (i) most of the participants were able to learn the approach in a limited time, (ii) all participants were able to list the advantages of using the approach; applying the approach helped to improve UML-based specifications. Due to the limited scope of the experiment (a single course with a limited number of participants), we plan to conduct further empirical studies to generalize the outcomes of the experiment. Nevertheless, the results demonstrate that the unified OO requirements approach has a potential to be adopted by requirements engineering practitioners.

References 1. Askarpour, M., Bersani, M.M.: Teaching formal methods: an experience report. In: Frontiers in Software Engineering Education: First International Workshop, FISEE 2019, Villebrumier, France, Nov 11–13, pp. 3–18. Springer, Berlim (2020) 2. Briand, L., Labiche, Y., Di Penta, M., Yan-Bondoc, H.: An experimental investigation of formality in uml-based development. IEEE Trans. Softw. Eng. 31(10), 833–849 (2005) 3. Briand, L.C., Labiche, Y., Madrazo-Rivera, R.: An experimental evaluation of the impact of system sequence diagrams and system operation contracts on the quality of the domain model. In: 2011 International Symposium on Empirical Software Engineering and Measurement, pp. 157–166 (2011) 4. Cohn, M.: User Stories Applied: For Agile Software Development. Addison-Wesley Professional (2004) 5. Cook, S., Bock, C., Rivett, P., Rutt, T., Seidewitz, E., Selic, B., Tolbert, D.: Unified Modeling Language (UML) Version 2.5.1. Standard, Object Management Group (OMG), Dec (2017) 6. Fricker, S.A., Grau, R., Zwingli, A.: Requirements engineering: best practice. In: Requirements Engineering for Digital Health, pp. 25–46. Springer, Berlin (2014)

Teaching Object-Oriented Requirements Techniques …

353

7. Frosch-Wilke, D.: Using uml in software requirements analysis-experiences from practical student project work. In: InSITE-Informing Science and IT Education Conference, pp. 175– 183 (2003) 8. Galinier, F.: Seamless development of complex systems: a multirequirements approach. Ph.D. thesis, Université Paul Sabatier-Toulouse III (2021) 9. Jacobson, I., Christerson, M., Jonsson, P., Övergaard, G.: Object Oriented Software Engineering: A Use Case Driven Approach. Addison-Wesley, Boston MA (1992) 10. Jacobson, I., Spence, I., Kerr, B.: Use-case 2.0: The hub of software development. Queue 14(1), 94–123, Jan (2016) 11. Khazeev, M., Aslam, H., de Carvalho, D., Mazzara, M., Bruel, J.-M., Brown, J.A.: Reflections on teaching formal methods for software development in higher education. In: Frontiers in Software Engineering Education: First International Workshop, FISEE 2019, Villebrumier, France, Nov 11–13, pp. 28–41. Springer, Berlin (2020) 12. Larman, C.: Applying UML and Patterns: An Introduction to Object Oriented Analysis and Design and Interative Development. Pearson Education India (2012) 13. Madrazo-Rivera, R.: An experimental evaluation of the impact of system sequence diagrams and system operation contracts on the quality of the domain model. PhD thesis, Carleton University (2007) 14. Meyer, B.: Applying “design by contract.⣞ Computer 25(10), 40–51 (1992) 15. Meyer, B.: Handbook of Requirements and Business Analysis. Springer, Berlin (2022) 16. Nadine Diaz-Infante, S.R., Michael Lazar, Ray, A.: Demand for online education is growing. are providers ready? McKinsey & Company (2022). https://www.mckinsey.com/ industries/education/our-insights/demand-for-online-education-is-growing-are-providersready Accessed on 10 Feb 2023 17. Naumcheva, M., Ebersold, S., Naumchev, A., Bruel, J.-M., Galinier, F., Meyer, B.: Objectoriented requirements: a unified framework for specifications, scenarios and tests. arXiv preprint. arXiv:2209.02189 (2022) 18. Wohlin, C., Höst, M., Henningsson, K.: Empirical research methods in web and software engineering. In: Web Engineering, pp. 409–430. Springer, Berlin (2006)

Onlife Education: Beyond Distance Learning by Intelligent Tutoring Systems Salvatore Distefano

Abstract In the pandemic scenario, characterized by lockdowns and service interruptions, Distance Learning allowed educational pathways remotely, proving to be a reliable solution, resilient to the emergency. However, issues related to its effectiveness as well as social and psychological implications, remain unresolved. From many sides, there is a desire to go beyond DL, towards a more engaging and efficient experience that takes advantage of the limitless knowledge resources offered by the infosphere. This new type of DL or DL 2.0 integrates different educational modes, beyond the blended approach, following the path of onlife education, without distinction between online and offline. In this context, tools such as Intelligent Tutoring Systems (ITS) could prove to be useful and effective as a complement to DL, a significant step towards onlife learning. In this paper, an ITS solution that led to the implementation of Virtual Study Buddy is proposed, primarily designed to support students in their individual studies, acting as a digital learning assistant. Virtual Study Buddy brings together concepts of machine learning and gamification with cloud technologies, leveraging personal and mobile devices, and can integrate with traditional DL processes for a new and more complete form of onlife learning.

1 Introduction COVID-19 pandemic limitations on personal interactions have led to an increase in web searches for the term “distance learning,” which has become a frequently used buzzword. Beyond the pandemic, distance learning, with all its pros and cons, remains a valuable tool for inclusiveness, to remove barriers for impaired, disable or vulnerable subjects, or even for those who live in remote or disadvantaged areas, opening the world of knowledge to all those who possess a device connected to the Internet. The approach to distance learning is subjective: with reference to the digital natives [1], the millennials, who are comfortable with technology and computers S. Distefano (B) University of Messina, 98122 Messina, Italy e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_33

355

356

S. Distefano

from a tender age, ICT is seen as an integral and necessary part of life. On the other hand, digital immigrants have mostly been swept away, unwillingly, by technology and show difficulties in effectively adapting to this world. However, it is important to note that most people, including digital immigrants, can learn and become proficient in using these technologies with some practice and training. There are many resources available to help people overcome these difficulties, such as training courses, online tutorials, and technology assistance programs. These programs can help individuals become familiar with digital technologies and make the most of them for their personal and professional needs. Ultimately, technology can be an opportunity for everyone, including digital immigrants, to stay connected, access information, and develop skills that can improve their lives. To assess the effectiveness of distance learning, social aspects need to be considered as part of the educational process. The development of interpersonal relationships among students such as friendship is one of the individual growth milestones, indeed. In addition to “leisure” encounters, the lack of shared study experiences outside the school context is a problem to be considered in distance learning. Prolonged periods of DL, remotely, without direct/physical contacts and interactions hinder the relationships. With whom to measure progress and limitations in learning, if not with our mates-friends? It is thus necessary to go beyond the current concept of distance learning and evolve it towards a more complete and immersive experience in the cyberspace and the infosphere. From a technical viewpoint, in times of social networks, virtual and extended reality, and metaverse, this is absolutely feasible. From a theoretical perspective, such ideas and related concepts are based on a theory recently emerged in a multidisciplinary (philosophical, sociological, anthropological, and technological) context led by Luciano Floridi (the philosophy of information father) [2] summarized with the word “onlife,” a contraction of “online,” “offline,” and “life.” Onlife is a neologism that identifies the vital, relational, social, and communicative, work, and economic individual dimensions as the result of continuous interactions between the real, analog life and the social network, virtual one. The onlife person lives a life that does not distinguish between online and offline, fluidly, in a physical-digital continuum. In the context of education, this concept can be expressed as “onlife learning” or more generally as “onlife education (OLE)”. OLE is a concrete attempt to go beyond the concept of emergency and distance learning towards a better and more integrated use of digital technologies in education, taking advantage of the positive aspects of distance learning and integrating it with offline tools to ensure a complete educational process without breaks and interruptions. In the onlife education, information and communication technologies break down the divide between online and offline, merged into onlife, and are no longer viewed as passive tools or resources. Rather, they are active mechanisms that enable the emergence of collective and networked intelligence, as well as educational ecosystems that shape the way we teach and learn [3]. The OLE concept moves towards “phygital” students, a neologism referring to the blending of physical and digital experiences and environments, often using technology to enhance physical experiences or incorporating physical elements into digital experiences. An OLE phygital students fluidly moves from physical/online to

Onlife Education: Beyond Distance Learning by Intelligent Tutoring …

357

digital/offline learning modes, sometimes even combining them to further improve the educational experience. In the process towards onlife education and DL 2.0 it is important to cover the full learning process, including the personal study. DL mainly covers the group training activities, supported by several tools, while personal, individual study are mostly left to the student. Intelligent tutoring systems (ITS) fill this gap by supporting the learning process performing similar functions to those of a human tutor, interacting with students in a natural and adaptable manner. Thanks to artificial intelligence, these systems can implement strategies for tutoring students. The ultimate goal of an ITS is to analyze the student competence and behavior within a digital learning system linked to a specific field of knowledge. The ITS can evaluate the difference between the student training situation and the training objectives to be achieved. Typically, during the training activity, the ITS provides students with appropriate comments and suggestions by selecting the most appropriate content and types of activity to help them correct themselves. Thereby, it fills the student gaps and allows them to progress in their individual educational process. Despite literature shows that ITS have improved student performance and learning, there are problems in their adoption. In addition to technical and organizational problems, ITS have shown aging problems of “longevity” in their use, usually due to a degradation of the student interest in interacting with such systems, which become mechanical and repetitive in the long run. An effective way to address this problem is gamification. The term gamification was coined by Nick Pelling in 2002, but it has become widely used only in 2010 [4]. Gamification consists of using game mechanics to influence performance, create responsibility and more generally influence the behavior of participants. These game mechanics satisfy some basic human psychological needs such as a sense of competence, autonomy, and relationship. Gamification uses “intrinsic motivation”, which is the strongest driver of long-term involvement, as well as sophisticated game mechanics and adopts a long-term approach to changing behavior and creating work habits for students. Through its power to communicate goals and provide real-time feedback on student results, gamification is an ideal tool for engaging and motivating students and is increasingly used in teaching. In this paper, an Intelligent Tutoring System (ITS), Virtual Study Buddy is proposed as a concrete step towards OLE, exploiting concepts of machine learning and gamification to support student learning processes. The goal is to provide a tool that can help students in the development of a proper study method by interacting with a virtual and digital study partner who can evaluate and suggest how to improve their performance, even learning from these interactions. Linking it to DL activities, the proposed tutoring system could be considered a first implementation of a tool for onlife education, ensuring an immersive experience in teaching, i.e. DL 2.0. In the following sections, the proposed solution is detailed, specifically providing preliminary concepts, including a brief overview of the literature references on the subject, in Sect. 2. The Virtual Study Buddy system, its features, and the usage scenario are described in Sect. 3. Section 4 concludes the work with a discussion of future developments.

358

S. Distefano

2 Preliminary Concepts As stated in [5], ICT can help create an educational system based on the principles of helping teachers, students, and administrations to be effective, improving the quality of the teaching–learning process. In the following, we will introduce the concepts of ITS, edutainment, gamification, through bibliographic references inspiring this work.

2.1 Background Is it worth using ITS? Despite significant research activity related to the development of ITS over the last 30 years, with many funded projects, a lot of money spent, and well-established experiments, the use of these systems has never been effectively launched, and it has never become systemic. The failure to use ITS in the real world outside university research laboratories is certainly not due to a lack of results obtained in the various experimental experiences described in the literature. All research suggests that intelligent tutoring systems can achieve significant increases in student learning compared to traditional educational communities. From a historical point of view, the research on ITS has the main aim of providing an advanced tutoring experience comparable to that obtainable with a human tutor rather than that obtained with traditional computer-assisted teaching (simple verification of the correctness of the answer given). From an operational point of view, many ITSs have not definitively entered the education system because they were difficult to manage. In most systems, “knowledge maintenance” had to be performed by expert programmers at great expense. This fact, in our opinion, has led to an increase in costs and times that instructors and educational institutions must consider, effectively blocking their diffusion and actual use. To avoid these problems, we decided to automate the knowledge maintenance phase using artificial intelligence techniques. These services can accept digital format teaching materials as input that does not require preprocessing (e.g., normal textbooks recommended in class). Therefore, based on the comparative process between these previously acquired knowledge and those provided orally and in real-time by individual users during subsequent learning sessions, the ITS can return evaluations on the acquired competencies and, therefore, to perform predictive analyses on future learning times. Why Gamification? According to [6], good video games are “learning machines” as they incorporate some of the most important learning principles postulated by modern cognitive science. In [7], the authors explain how a good gamification process requires the presence of two essential components: effective incentivization dynamics and proper technology, considering “gamification is 75% psychology and

Onlife Education: Beyond Distance Learning by Intelligent Tutoring …

359

25% technology.” From a psychological point of view [8], it is possible to identify three fundamental phases to effectively involve game participants: • Motivating. The starting point of every gamification activity is to give people a reason to participate. The game and challenge mechanism are deeply rooted in the human mind and are a powerful stimulus, but to make it work best, it is essential that players have a reward, a goal, an objective that attracts attention and increases determination. The choice of benefits and rewards is essential because the more accurate it is, the greater the competitive drive that will be generated in the group. • Provide tools to participate. For gamification to work, it is necessary to provide participants with the tools to participate. The interface must be intuitive and easy to use, allowing everyone to interact and perform the tasks required to achieve the objectives, achieving a tradeoff between simplicity and effectiveness. • Kickoff. Gamification activities require a moment to establish a starting point, such as an event, team building activity, or official communication. Long-term competitions should have intermediate milestones to monitor progress, reward participants, celebrate successes, and motivate those who are struggling. Timing is crucial, as activating game mechanics simultaneously and in a coordinated manner is essential to retain the participant interest.

2.2 Related Work Edutainment is a form of entertainment that is designed to educate its audience about specific subjects or concepts while engaging them in an entertaining and interactive manner. The concept of edutainment has been applied to a wide range of mediums, including video games, television shows, movies, and live events. Edutainment can be particularly effective in the areas of science, technology, engineering, and mathematics (STEM) education, as well as for promoting health behaviors and environmental awareness [9–11]. One of the key elements of effective edutainment is the incorporation of game-like elements, such as points, rewards, and levels, into the learning experience [12]. This approach is based on the principles of gamification, which involves the application of game design elements to non-game contexts to motivate and engage individuals in specific behaviors [13]. With specific regard to the ITS, the use of gamification has been explored in several studies. In [14], the authors present empirical results on teaching basic Mandarin as a second language to university students using a gamification approach. In [15], the authors describe how the game “Musou Roman” can help Japanese culture enthusiasts learn the more complex aspects of the language through gamification. In [16], the authors explore key elements that can lead to successful gamification in the context of history as a learning context, guided by student motivation and based on the Octalysis framework. In [17], the authors describe and analyze some gamification methods used by the Zagreb School of Economics and Management in various technology and legal discipline courses. In [18], primary school students used the Octalysis structure proposed in [19] for educational practices. In [20], the authors analyze the application

360

S. Distefano

Fig. 1 Virtual study buddy ITS high-level scheme

of gamification strategies in MOOCs. In [21], the authors present a didactic strategy that integrates gamification with traditional teaching methods for final-year students in Computer Science and Engineering. In [22], the authors successfully applied gamification to the study of the Quran to improve learning. In [23], the authors conducted an exploratory study to evaluate the effect of using gamification through interactive digital storytelling on classroom dynamics and student interaction.

3 Virtual Study Buddy Given the above, the proposed research endeavors to create a computational system that aids students in their educational pursuits. To this end, the design and development of an Intelligent Tutoring System Android app, Virtual Study Buddy (VSB), currently available in a beta version,1 as has been performed. Our prototype and freely accessible to users who have registered with their Google or Facebook accounts. The architecture of Virtual Study Buddy, depicted in Fig. 1, is primarily composed of the mobile application, Google Cloud Platform services,2 and a Firebase database3 exploiting JSON files to communicate via HTTP. The client implemented by the app allows students to access the VSB ITS, which can detect, and process data related to user learning activities regarding a subject or concept submitted as written text, thus providing assessments of acquired skills. As illustrated in Fig. 1, the application operates in two distinct modes: the Basic Knowledge Creation Mode, which acquires the topic and foundational knowledge to be studied, and the Training Mode, which assesses the knowledge obtained by the student. The workflow of Virtual Study Buddy, encompassing all phases of the tutoring process, is depicted in Fig. 2. The modular architecture of the system, which 1

https://play.google.com/store/apps/details?id=com.knowledgepkg.domenicogiacomocampanile. knowledgeapp. 2 https://cloud.google.com/. 3 https://firebase.google.com/.

Onlife Education: Beyond Distance Learning by Intelligent Tutoring …

361

Fig. 2 Virtual study buddy ITS workflow

Fig. 3 Virtual study buddy architecture

includes third-party services, is shown in Fig. 3. The two VSB modes are detailed in the following.

3.1 Basic Knowledge Creation Mode In the Basic Knowledge Creation Mode, after the login phase, the application prompts the user to input textual content. This content is classified by associating it with a topic and subtopic during the creation phase. This textual content represents what we have called foundational knowledge, that is, the reference text upon which all future analyses and considerations will be based (further details will be provided later). For example, we have decided to create a new foundational knowledge called “Alexander the Great”. We have assigned it to the history topic and the biography subtopic. The app allows for the content input in three different ways (as shown in Fig. 4): (i) by typing it directly into the text field using the Android keyboard (Fig. 4a); (ii) by

362

S. Distefano

a - text

b - image

c - pdf

Fig. 4 Virtual study buddy: basic knowledge creation input example

capturing it from an image (or photo) (Fig. 4b); and (iii) by extracting it from a PDF file (Fig. 4c). The file from which to extract the text can be located either on the device or on an external cloud storage service. Once the text is acquired and confirmed by the student/user, it is sent and processed by the online services to extract text and metrics, which are then stored in the Firebase database. These metrics will be used for the assessment at the end of each Training Mode session.

3.2 Training Mode In the Training Mode, the app prompts the student to select a specific previously entered basic knowledge to study and repeat, as shown in Fig. 5. The training consists of recording the student oral exposition on that basic knowledge, which is done using Google Cloud Speech to Text (S2T).4 Students can repeat their basic knowledge in a personal way, not strictly identical to the text they previously uploaded to the system, as semantic analysis is performed. The analysis examines the similarity, conformity, oral exposition, and similarity of meaning between the text entered during the Basic Knowledge Creation Mode and the text from the transcription of the vocal received during the Training Mode. The two texts are compared and analyzed structurally, syntactically, and semantically. 4

https://cloud.google.com/speech-to-text/.

Onlife Education: Beyond Distance Learning by Intelligent Tutoring …

363

Fig. 5 Virtual study buddy: training and assessment

The system analyzes every part of speech, detecting the morphology, dependency on other words present, and the taxonomy of the text to match the topics, concepts, and words present in both texts. At the end of the analysis, the app stores the metrics related to this training session in the Firebase DB. Then, it compares text and metrics of the basic knowledge with text and metrics of the current training session. The results are stored in the DB and shown to the user on the app (Fig. 6). The score is expressed in hundredths, taking into account: (i) oral exposition; (ii) equivalence of texts; (iii) similarity of the topic; (iv) percentage of acquired text knowledge; (v) time spent to rehearsal.

4 Conclusions This paper introduces an Intelligent Tutoring System based on machine learning and gamification concepts, with the main goal of supporting students in the educational process. Virtual Study Buddy has been released to support Distance Learning. By integrating VSB into a DL tool a significant step towards onlife education, blending online and offline education practices, can be implemented. Future framework implementations will move in this direction, increasing integration with current DL tools on one hand, and further automating interaction with students on the other, for example by automatically generating questions to evaluate their preparation. To support such DL 2.0, augmented, virtual extended and mixed reality technologies, the metaverse and social networks can be exploited to implement the onlife, phygital dimension of a brand new generation of students.

364

S. Distefano

Fig. 6 Virtual study buddy: learning statistics

References 1. Prensky, M.: Digital natives, digital immigrants part 1. On Horizon 9(5), 1–6 (2001). https:// doi.org/10.1108/10748120110424816 2. Norberg, A.: From blended learning to learning onlife: ICTs, time and access in higher education. Ph.D. thesis, Umea University (2017) 3. Floridi, L.: The onlife manifesto: being human in a hyperconnected era. Springer Nature (2015) 4. Kamasheva, A., Valeev, E., Yagudin, R., Maksimova, K.: Usage of gamification theory for increase motivation of employees. Mediter. J. Social Sci. 6(1 S3) (2015). http://www.mcser. org/journal/index.php/mjss/article/view/5674 5. Keswani, B., Banerjee, D., Patni, P.: Role of technology in education: a 21st century approach. J. Commerce Inform. Technol. 8, 53–59 (2008) 6. Gee, J.: What video games have to teach us about learning and literacy. Comput. Entertain. 1, 20 (2003). https://doi.org/10.1145/950566.950595 7. Zichermann, G., Cunningham, C.: Gamification by Design: Implementing Game Mechanics in Web and Mobile Apps. O’Reilly Media, Inc. (2011) 8. Fogg, B.: Persuasive technology. In: Fogg, B. (ed.) Persuasive Technology. Interactive Technologies, Morgan Kaufmann, San Francisco (2003). https://doi.org/10.1016/B978-1558606432/50001-9 9. Gee, J.P.: Good video games and good learning. Phi Kappa Phi Forum 85(2), 33–37 (2005) 10. Giddings, S.: Edutainment: designing educational video games for remote Australian aboriginal communities. In: Proceedings of the 2015 Annual Symposium on Computer-Human Interaction in Play, pp. 783–785 (2015) 11. Joseph, B.: Edutainment and effective learning. J. Educ. Social Res. 5(3), 195–203 (2015) 12. Dondlinger, M.J.: Educational video game design: a review of the literature. J. Educ. Technol. Dev. Exchange 1(1), 1–19 (2007) 13. Deterding, S., Dixon, D., Khaled, R., Nacke, L.:. From game design elements to gamefulness: defining “gamification”. In: Proceedings of the 15th International Academic MindTrek Conference: Envisioning Future Media Environments, pp. 9–15 (2011)

Onlife Education: Beyond Distance Learning by Intelligent Tutoring …

365

14. Heryadi, Y., Muliamin, K.: Gamification of m-learning mandarin as second language. In: 2016 1st International Conference on Game, Game Art, and Gamification (ICGGAG), pp. 1–4 (2016). https://doi.org/10.1109/ICGGAG.2016.8052645 15. Fathoni, A.F.C.A., Delima, D.: Gamification of learning kanji with “Musou Roman” game. In: 2016 1st International Conference on Game, Game Art, and Gamification (ICGGAG). pp. 1–3 (2016). https://doi.org/10.1109/ICGGAG.2016.8052664 16. Ymran, F., Akeem, O., Yi, S.: Gamification design in a history e-learning context. In: 2017 International Conference on Information, Communication and Engineering (ICICE), pp. 270– 273 (2017). https://doi.org/10.1109/ICICE.2017.8479194 17. Aleksic-Maslac, K., Rasic, M., Vranesic, P.: Influence of gamification on student motivation in the educational process in courses of different fields. In: 2018 41st International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO), pp. 0783–0787 (2018). https://doi.org/10.23919/MIPRO.2018.8400145 18. Cunha, G.C.A., Barraqui, L.P., de Freitas, S.A.A.: Evaluating the use of gamification in mathematics learning in primary school children. In: 2018 IEEE Frontiers in Education Conference (FIE). pp. 1–4 (2018). https://doi.org/10.1109/FIE.2018.8658950 19. Chou, Y.k.: Actionable gamification: beyond points, badges, and leaderboards. Packt Publishing Ltd (2019) 20. Romero-Rodriguez, L.M., Ramìırez-Montoya, M.S., González, J.R.V.: Gamification in moocs: engagement application test in energy sustainability courses. IEEE Access 7, 32093–32101 (2019). https://doi.org/10.1109/ACCESS.2019.2903230 21. Dixit, R., Nirgude, M., Yalagi, P.: Gamification: an instructional strategy to engage learner. In: 2018 IEEE Tenth International Conference on Technology for Education (T4E). pp. 138–141 (2018). https://doi.org/10.1109/T4E.2018.00037 22. Rosmansyah, Y., Rosyid, M.R.: Mobile learning with gamification for alquran memorization. In: 2017 International Conference on Information Technology Systems and Innovation (ICITSI), pp. 378–383 (2017). https://doi.org/10.1109/ICITSI.2017.8267974 23. Molnar, A.: The effect of interactive digital storytelling gamification on microbiology classroom interactions. In: 2018 IEEE Integrated STEM Education Conference (ISEC), pp. 243–246 (2018). https://doi.org/10.1109/ISECon.2018.8340493

A Personal View on Past and Future Higher Education Nikola Zlatanov

Abstract In this paper, I first provide a brief overview on the history of institutional educational systems and show that universities had to evolve in order to cope with the challenges of the time. One such major challenge, which requires another round of evolution of the university, is online education. To understand the gravity of this challenge to the university, I discuss the advantages and the disadvantages of online education from the university’s and students’ perspective. Finally, I provide some predictions for the future of online higher education in the area of information technology (IT).

1 A Brief History of Education The word education comes from the word educare, which means to bring up (a child) [1]. The etymology of the word education is aligned with the main purpose of education, which is to perpetuate the accumulated knowledge from the current and past generations onto the next generation. The main purpose of early forms of education systems were to perpetuate the religious beliefs from the current and past generations onto the next generation. Based on current findings, there existed organized education in Egypt around 3500 BC [2]. One of the main roles that this early educational system conducted was to perpetuate the Egyptian writing system (hieroglyphs) from the current generation onto the next. Educational systems also appeared in other regions of the world. For example, during the pick of Athens’s power, the famous schools of Aristotle, Plato, and Socrates were opened where students were thought philosophy [3]. Some of these schools lasted several centuries. The main invention that these ancient schools brought to the public was the broadcast technology, which is the ability of one person to lecture to dozens/hundreds of students. Students that attended these ancient schools had to have sufficient leisure time that could then be allocated to attending school and education. In fact, the word school comes from the Greek word for leisure, in the sense N. Zlatanov (B) Innopolis University, Innopolis, Russia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_34

367

368

N. Zlatanov

that students had to dedicate their leisure time to attend schooling [1]. Consequently, those that could not afford any leisure time, could not join the schooling process. The phenomenon of dedicating one’s leisure time to schooling is present even today. Later in Roman times, the three-phased educational system was invented, comprised of an early forms of primary, secondary, and tertiary education [4]. Primary schools taught literacy and basic mathematics, secondary taught literature and arts. Finally, tertiary education in Rome, accessible only to the elites, was concerned with teaching oratory, law, and philosophy with the aim of preparing the students to become the future political elites that would run the county. The elitism of tertiary education prevailed, in a modified form, almost to the 16th century and beyond. In the begining of the middle ages, formal education was mostly performed by religious organizations [7]. As a result, in Europe, the main educational organization, in terms of quantity, was the Church. Students attended religious educational organizations in order to join the clergy. As time progressed, due to necessity, other educational organizations appeared aimed at educating those students that were not able to join the religious educational organizations [5]. These non-religous educational organizations quickly grew in numbers and attracted high numbers of students and teachers. At some point, the teachers working in these educational organizations formed unions in order to maintain high standards of teaching, as well as to preserve their interests against outside forces. One such union was the universitas magistrorum et scholarium, which means union of teachers and scholars, which is the origin of the name university as well as the origin of today’s university administrative structure. Even in the middle ages, the universities had different faculties teaching Arts, Medicine, Law, and Theology. Moreover, similar to today’s university rankings, in the middle ages different universities in Europe were famous for different faculties [6]. For example, Paris was famous for Arts and Bologna was famous for Law. The curriculum of a university course was composed of a list of texts and books [6]. Every faculty used books of authors with well established competences in the area from which they taught the courses. The university professors had to convey to the students the contents of those books without any alteration. The lectures were divided into two parts. In the first part, the professor would read the text to the students and in the second part the students would discuss the content of what was read by the professor. This teaching method invented in the medieval ages continued up to this date, with some modifications. Graduating from these universities was not easy nor cheap [6]. For example, a Law student at the University of Bologna could reach the final examination for graduation only after 8 years of study. The first phase of the final examination would involve an interview with a university professor, where the student would be questioned on the entire subject. If this stage is passed, in the next and even harder phase, the student will face the entire faculty. Only if the student passes the second phase, he would go to the third and final phase, the public phase, which was very costly. Of course, this lengthy and costly university educational process created elitism, since only a few students could attend and graduate successfully. The elitism in the tertiary education, which started as early as Roman times and continued through the middle ages, started to fade out in the 16th century, when the colonial European states required higher

A Personal View on Past and Future Higher Education

369

numbers of university graduates that could be sent to administer state functions in the newly acquired colonies. The above brief overview of institutional education shows how educational institutions evolved over time. This evolution was due to different challenges that educational institutions faced, such as unionizing in order to protect their interests, to restructuring in order to increase the number of graduates when that was required by the colonial states. Each time, the educational institutions adapted and reorganized in order to keep with the modern trends of the time. In fact, the current tertiary educational system is simply an adaptation of the middle age universities to modern times. At this point in time, the university is again facing a major challenge due to the penetration of the Internet and its consequences.

2 Current and Future Education The university educational system that we had until the wide penetration of the Internet was very similar to the university educational system established in the middle ages. One of the main differences between the two is that many more students could attend university, hence, the amount of elitism compared to the middle ages is significantly reduced. In terms of the teaching method, similar to the middle ages, each student had to come to the university campus, physically attend the lectures of the professor, physically attend labs/tutorials based on the lectures, and at the end pass the exam. Hence, that main constraint of the university educational system, up to recently, was the compulsory university onsite attendance by the students. This constraint was not forced by the university, instead, it was due to the natural order of how a student could hear and watch the lectures of a professor. With the penetration of the Internet, this constraint become obsolete, since now a student could hear and watch the lectures of a professor online [8]. Although online teaching is welcomed by the university, removing the students’ compulsory university onsite attendance, is not. This is because almost all universities have invested huge amounts of funds into building glamorous university campuses that would attract higher number of students. Hence, having a purely online university education would mean that these campuses would become vacant and thereby obsolete [9]. As a result, the huge amounts of funds invested into the campuses become would become a malinvestment. Another reason why universities resist complete online higher education is increased competition. Specifically, up to recently, the university campus enabled a university to cover a specific geographical area from which it could recruit students. Competition with other universities in that geographical area, to a large extend, was significantly reduced. Moving to a completely online education would result in the universities loosing the geographical edge. In that case, students would not be compelled to go to the closest university and could chose any online university independent of the location. Hence, completely online education would create a fierce

370

N. Zlatanov

competition among the universities leaving the low ranked universities in a highly disadvantaged position compared to the high ranking universities. From the student’s perspective, complete online education is highly valued since each student can study at their own pace without leaving their home [10]. However, the main disadvantage of complete online education is the inability of students to make connections with other students [11]. During onsite campus presence at faculty buildings or at student dorms, students are able to interact physically and thereby create connections that would last and benefit them for years. Such strong connections are almost impossible in the virtual reality. The future of completely online university was experienced during the government imposed lockdowns as their response to the Covid-19 pandemic. During the lockdowns, the university campuses had to be emptied and therefore the entire educational process had to move strictly online. Due to the empty campuses, the universities saw their revenues decline and their expenses either stayed the same or increased. As a result, many of the universities faced bankruptcies and turned to laying off staff in order to cope with their budget deficits. Difficulties were especially experienced by universities that did not have any financial support from governmental tax revenues, and were low ranked. However, from the lockdown experience, both the students and the teaching staff realized that complete online learning/teaching is possible for most subjects, at least in the information technology (IT) area. In fact, students and their professors in the IT area adapted to strictly online teaching to such an extent that the university administration had to force both the professors and the students to compulsory onsite attendance of courses. Currently there is an ongoing discussion among university administrators on how the university should adapt to the online learning challenges and their consequences. Many solutions are proposed and discussed. Finally, some solutions will emerge and with their implementation the current higher educational system would be adapted to a one that is suitable for the needs of the future. It is very hard to provide predictions of what these solutions would be. However, it is clear that for areas such as IT, where all of the lectures, labs, and tutorials can be completely provided online, the future is completely online education with some form of a reach in-person experience.

3 Conclusion Universities have adapted to challenging situations and evolved accordingly. The post lockdown period showed that completely online university education is not only doable, especcialy in the IT area, but much more efficient in terms of learning than onsite university education. However, the main disadvantages are empty campuses, the lack of strong connection between students (and staff) that are formed during onsite campus attendances. Hence, universities have to adapt and find solutions that would enable them to jointly provide online learning and reach in-person experiences at their campuses.

A Personal View on Past and Future Higher Education

371

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.

Harper, D.: Online Etymology Dictionary (2001) Marrou, H.: A History of Education in Antiquity. University of Wisconsin Press (1982) Hibler, R.W.: Life and Learning in Ancient Athens. University Press of Amer (1988) Bonner, S.: Education in Ancient Rome: From the Elder Cato to the Younger Pliny. Routledge (2012) Rashdall, H.: The Universities of Europe in the Middle Ages, vol. 1. Clarendon Press (1895) Briggs, A.: A History of the University in Europe: Volume 1, Universities in the Middle Ages, vol. 1. Cambridge University Press (1992) Thorndike, L.: Elementary and secondary education in the middle ages. Speculum 15(4), 400– 408 (1940) Neuwirth, L.S., Jovi´c, S., Runi Mukherji, B.: Reimagining higher education during and postCOVID-19: challenges and opportunities. J. Adult Continuing Educ. 27(2), 141–156 (2021) Pickerel, S., Chhetri, R.: A literature review on impact of COVID-19 pandemic on teaching and learning. High. Educ. Future 8(1), 133–141 (2021) Rashid, S., Yadav, S.S.: Impact of Covid-19 pandemic on higher education and research. Indian J. Human Dev. 14(2), 340–343 (2020) Aristovnik, A., et al.: Impacts of the COVID-19 pandemic on life of higher education students: a global perspective. Sustainability 12(20), 8438 (2020)

Running Regular Research Seminar Online N. V. Shilov , D. A. Kondratyev , N. Kudasov , and I. S. Anureev

Abstract We present profile and experience of a hybrid seminar on fundamental issues of software engineering, theory and experimental programming ru-STEP (= russian seminar on Software Engineering, Theory and Experimental Programming) during 30 months of its history. The seminars was launched 3 months before COVID19 outbreak, worked during two years of COVID-19, survived and continues its work after COVID-19 restrictions were dropped. We do not attempt to provide a comprehensive study or systematic review of world-wide experience to run online research seminars (especially in time of COVID-19) that need more studies and could be a topic for later research.

1 ruSTEP: Over COVID-19 Years and Beyond Hybrid (offline and online) seminar ruSTEP on fundamental issues of software engineering, theory and experimental programming was launched on Wednesday October 21, 2020, just three months ahead of COVID-19 pandemic outbreak. Acronym ruSTEP stays for Russian seminar on Software Engineering, Theory and Experimental Programming. The title of the seminar and its acronym were coined by Aleksandr Naumchev, one of founders of the seminar and member of the Steering Committee. Complete list of the Steering Committee may be found on ruSTEP webpages in English at https://persons.iis.nsk.su/en/ruSTEP and in Russian at https:// persons.iis.nsk.su/ru/ruSTEP.

N. V. Shilov (B) · N. Kudasov Innopolis University, Innopolis, Russia e-mail: [email protected] N. Kudasov e-mail: [email protected] D. A. Kondratyev · I. S. Anureev A.P. Ershov Institute of Informatics Systems, Novosibirsk, Russia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_35

373

374

N. V. Shilov et al.

The seminar was (and still is) an initiative of the Steering Committee—a group of researchers from Innopolis University https://innopolis.university/en/ and of A.P. Ershov Institute of Informatics Systems (https://www.iis.nsk.su/en). The primary purpose of the Committee was to create and maintain an online platform and venue for regular (bi-weekly) professional meetings of researchers in Russia to present and discuss their studies in the field of fundamental issues of software engineering, theory and experimental programming, curricular and teaching experience (in the related fields). A bi-lingual mode (reflected in the bi-lingual web-pages) was motivated by growing presentation of international scientists in Russian research institutions as well as by openness to the world, Urbi et orbi. It is recommended to prepare a presentation in English, to give a talk and run discussion in Russian (if there are no international participants), but be ready to switch to English (in the case of international audience). In total there were more than 56 seminar sessions (+4 sessions for Russian audience). Links to the presentations (in the pdf-format) of a majority of the talks (since February 5, 2021) are provided on both official web-pages. Seminar’ videoarchive (since January 15, 2021) is available as a playlist on IIS YouTube-channel https://clck.ru/33VMc5. Seminar has 36 subscribers for emails advertising next meetings/sessions. The actual attendance/participation in the meetings ranged from 3 (speaker, moderator and a listener) to 28 (in COVID-19 times). As it follow from the starting date of ruSTEP (October 21, 2020), the Steering Committee did not assume any emergency situation in science communications as caused by anti-COVID measures or political sanctions (because of the special military operation in Ukraine). Our paper is not appropriate place to discuss causes and consequences of the listed global events, we just would like to report our experience during these stream of events. Please be aware that some experience may be cause by these events, while some experience reflects processes and situation in academia, research, and education. Steering Committee had intention to form and advertise (via web-pages and emails to subscribers) seminar agenda a month in advance, but actually quite frequently upcoming meetings had been advertised just 3–5 days before the meeting (while biweekly periodicity was maintained). Unfortunately, our hope (explicitly written on the web-pages) that volunteers to give a talk will contact any/several/all member(s) of the Steering Committee with talk proposals never come true during the 30 month history of the seminar. A Google-group was created to discuss seminar-related issues and questions raised in the talks and presentations. The only time a short discussion arose and was conducted in the group; maybe, we should move to more contemporary communication tools (like chats in messengers)? As it was mentioned above, 36 participants registered on the mailing list (through a Google form), but, unfortunately, just few from presenters invited by Steering Committee registered in the list and but dropped out until the next invitation to give a talk. Maybe it makes sense to advertise seminar meeting not only via our own mailing list but on other topic-oriented mailing lists (e.g., Types-list https://lists.seas.

Running Regular Research Seminar Online

375

upenn.edu/mailman/listinfo/types-list) and portals for online-seminar advertising (e.g., https://researchseminars.org/). Unfortunately (— indeed!), the seminar was integrated/included with/to the curriculum and educational process neither of Innopolis University nor Novosibirsk State University (https://english.nsu.ru/) in spite of strong engagement of A.P. Ershov Institute of Informatics Systems with Novosibirsk State University. Nowadays master and Ph.D. students have a heavy study, research, and (frequently) teaching load that prevents many of them from regular and active involvement into extra-curricular research seminar activities. We believe that Steering Committee should work towards inclusion/integration of the seminar to graduate studies curricular in both universities. Seminar meetings run fully online during the COVID-19 sanitarium measures (lockdown), but before and after this period some meetings were held in a mixed format at Innopolis University (and never at Institute of Informatics Systems in Novosibirsk). Firstly we used Zoom for online meetings using either university accounts or (later) private ones (while it was possible to pay for an annual subscription using international payment system), and since August 2022—(free) Skype. It turns out that online audience of ruSTEP doesn’t like Skype (even it can be used as a browser application) while organizers see just a single inconvenience (automatic call to all who have ever attended any previous meeting of the seminar in Skype). Several times the planned days of the meetings coincided with the meetings of conferences, workshops, public lectures on topics related to the topics of the seminar. In these cases the seminar published an invitation to go to visit (i.e., to attend these events online): Workshop VeriDevOps Project, Andrey P. Ershov memorial lectures, annual Spring/Summer Young Researchers’ Colloquium on Software Engineering, Open Conference of the Institute for System Programming of the Russian Academy of Sciences, and Workshops on Program Semantics, Specification and Verification https://persons.iis.nsk.su/en/PSSV21. Steering Committee believes that it is a good practice to make explicit respect to relevant annual events rather then to compete with them. Another interesting experience is a tradition to hold special meetings dedicated to the World Logic Day (https://en.wikipedia.org/wiki/World_Logic_Day (January 14), International Women’s Day (March 8), Seminar’s Problem Day (usually December 30), in memory of the Teachers—Igor V. Pottosin (1933–2001), Valery A. Nepomniachtchi (1939–2021). Steering Committee thinks that it is a good experience and should become a tradition. As it was stated above, it total ruSTEP had 60 meetings. Among them we had 6 reports representing the results of Ph.D. theses, 14 reports of Innopolis University employees, and 9 reports of employees of Institute of Informatics Systems. Researchers from Russia (namely, from Barnaul, Irkutsk, Krasnoyarsk, Moscow, Novosibirsk, Pereslavl-Zaleskiy, St. Petersburg, Saratov), from India, the United States, Switzerland gave talks at the seminar. By the way, some international presentation have been done offline at Innopolis University. Unfortunately, after start of the special military operation in Ukraine, our attempts to invite speakers from European Union have failed (candidate presenters that had promised to give a talk have token their promises away). Maybe, this change of

376

N. V. Shilov et al.

minds of our colleagues was more important for ruSTEP than COVID-19 story. Steering Committee is very sure that moves like these of our international colleagues are absolutely wrong and counterproductive: researchers should work together for peace and sustainability together worldwide. At least, ruSTEP is intended to continue to running regular research seminar in hybrid mode for Russian and International audience in year 2023. Later we plan to classify by topics all ruSTEP’ talks, meetings, and events, and to survey selected talks and events in more details. But in this conference paper (because of the space limitations) let us survey (in the next section) just talks in the field of program/system analysis, semantics, specification, and verification.

2 Selected Contributions on Analysis, Semantics, Specification, and Verification Talks on program/system analysis, semantics, specification, and verification presented on ruSTEP may be grouped (preliminary) into the following topics: • • • • • • • • • • • •

specification approaches, ways to define semantics of programming languages, verification conditions simplification, model reduction for efficient model checking, type-checking and inference, static analysis approaches and systems, verification of safety properties, verification of concurrent systems, verification of cyber-physical systems, verification of multi-agent system, proving verification conditions, error localization using formal methods.

Different approaches to formal specifications were demonstrated in talks by Aleksandr Naumchev, by Dmitry Kondratyev, and by Natalia Garanina. Aleksandr Naumchev suggested to use design-by-contract method to solve this problem. This approach is based on importance of defining program contracts during software development life cycle. The approach is to define seamless requirements and specifications for program components and its interfaces. This approach allows developers to implement more reliable and self-documented programs [22]. Talk by Dmitry Kondratyev contains an approach to loop invariants generation for loops with finite iterations over data sequences. This approach is based on replacement of variables from definite iterations by special recursive functions in verification conditions [23]. This replacement allows to avoid user-defined loop invariants in the case of finite iterations [14]. Natalia Garanina presented approach for defining Event-Driven Temporal Logic requirements. This logic allows verifying temporal and timed Properties of reactive

Running Regular Research Seminar Online

377

software using model checking. The approach is designed to provide developers by a set of templates of formulas that are often used in specifications [31]. Representation of these templates in natural language allows users to simplify specification process. Ways to define semantics were addressed in talks by Igor Anureev, by Dmitry Kondratyev, by Violetta Sim, and by Vasil Dyadov. Igor Anureev suggested an ontological approach to the deductive verification of imperative programs. The approach is based on a new type of transition systems— attribute systems, the specification language of attribute systems ASL and new types of semantics of imperative programming languages—ontological, ontologicaloperational, and ontological-logical [1]. This approach was illustrated using a model imperative language. Dmitry Kondratyev proposed to use method of verification conditions’ metageneration. This method allows to use an axiomatic semantics of a programming language as special input to verification system [21]. A special language for defining axiomatic semantics has been developed for this [12]. Violetta Sim presented ϕ-calculus (joint research with Nikolai Kudasov). The calculus is an untyped calculus of decorated objects [15], it allows to define semantics for object-oriented programming languages. Vasil Dyadov described use of K-framework for defining operational semantics. K-framework is based on Matching Logic [26]. Its sentences, the patterns, are constructed using variables, symbols, connectives, and quantifiers. In models, a pattern evaluates into a power-set domain (the set of values that match it), in contrast to firstorder logic where functions and predicates map into a regular domain (The more precise description of matching logic was presented in a survey by Nikolay Shilov later). Approach to verification conditions simplification depends on programming language semantics and relies upon a preliminary static analysis. The problem has been discussed in talks by Dmitry Kondratyev and by Mariya Ushakova. Talk of Dmitry Kondratyev contains description of mixed axiomatic semantics method. This method is based on using certain inference rules in the case of certain program constructs [18]. This approach allows applying more simple memory model in the case of so-called Pascal-like variables (reference and dereference operators are not applied to this variables). Static analysis can find such variables in some cases. Pifagor programming language presented by Mariya Ushakova is based on the data driven functional parallel computing model. This model allows considering only logic of data flow dependencies in the program [30]. Thus, this model simplifies the process of formal verification. Problem of reduction of state-space for simplifying model checking was addressed in talks by Sergey Staroletov, by Julio Cesar Carrasquel, and by Nataliya Garanina. Sergey Staroletov proposed special strategies for simplifying model checking, for example, using two 32-bit integers to model 64-bit integer instead of using long arithmetic (big integers [29]. Julio Cesar Carrasquel presented use of nested Petri nets instead of use of classic Petri nets. This approach allows reducing some states at top level [5].

378

N. V. Shilov et al.

Talk of Nataliya Garanina contains suggestions of using bounded model checking. Proposed constraints allows checking less states [31]. Type checking and type inference may be considered as important part of formal verification in some cases. Type-based approaches were considered in talks by Andrei Klimov, by Vitaly Romanov, and by of Nikolai Kudasov. Andrei Klimov described a dependent type theory with abstractable names proposed by Pitts [24]. Vitaly Romanov presented combination of CodeBERT and Graph Neural Networks for type prediction in the case of dynamically typed programming language Python [25]. Nikolai Kudasov presented the possibility of representing various higher-order unification problems as a special case of E-unification for second-order algebra of terms. This allows presenting beta-reduction rules for various application terms, and some other eliminators as equations, and reformulate higher-order unification problems as E-unification problems with second-order equations [16]. Nikolai Kudasov described also a procedure for deciding such problems by introducing second-order mutate rule (inspired by one-sided paramodulation) and generic versions of imitate and project rules. He demonstrated a prototype Haskell implementation for syntax-generic higher-order unification based on these ideas in Haskell programming language. Verification of safety properties was considered in talks by Natalia Garanina and by Alexander Kogtenkov. Talk of Natalia Garanina contains approaches to verification safety properties in Event-Driven Temporal Logic [31]. Alexander Kogtenkov presented approach of verification such safety property as void-safety which provides a guarantee of absence of dereferencing null-pointer (or void-reference) [10]. He presented checking correctness of one of the published approaches to guarantee of void-safety (Freedom before commitment by Summers and Müller). Alexander Kogtenkov had shown through a formalization in Isabelle/HOL that this approach contains errors and then has corrected them. Formal verification of concurrent systems is an important problem because of complexity of testing of such systems. The problem was discussed in talks by Natalia Garanina, by Julio Cesar Carrasquel, and by Artem Burmyakov. As was mentioned in the above, talk by Natalia Garanina contains suggestions of verification of concurrent systems using Event-Driven Temporal Logic. Julio Cesar Carrasquel proposed using nested Petri and classic Petri nets to verify concurrent systems [5]. Artem Burmyakov described consensus problem and available solutions [4]. In particular, Artem Burmyakov demonstrated applications of the consensus problem to compare the synchronization power of various concurrent data structures (e.g., a FIFO queue) and CPU synchronization primitives for the development of concurrent non-blocking algorithms, by means of consensus numbers. Verification of cyber-physical systems was discussed in talks by Natalia Garanina and by Sergey Staroletov. Sergey Staroletov presented use of Promela language for encoding tasks and SPIN verifier for checking obtained models [29].

Running Regular Research Seminar Online

379

Multi-agent system verification has been considered in talks by Natalia Garanina and by Nikolay Shilov. Natalia Garanina has proposed to use ontological approach for solving this task [8]. Talk of Nikolay Shilov contains description of Mars robot puzzle solution (a multiagent approach to the Dijkstra problem) [3]. Proving verification conditions is an important problem for practical application of formal verification. This problem was presented in talks by Dmitry Kondratyev and by Kirill Ziborov. Talk of Dmitry Kondratyev contains strategies for proving automation using ACL2 theorem prover [11, 13]. The main ACL2 feature is automation of proving by rewriting (based on using lemmas) and automation of proving by induction [20]. Kirill Ziborov reported experience of use of HOL4 theorem prover [28] and auxiliary lemmas for proving smart contracts [17]. Problem of error localization using formal methods was presented in talks by Dmitry Kondratyev, by Andrey Belevantsev, and by Nikolay Shilov. Dmitry Kondratyev proposed method of error localization and its implementation in the C-lightVer verification system. This approach contains method of generation of explanations in natural language about correspondence between subformulas of verification conditions and program fragments [7]. Also this approach contains strategies of error localization [14]. These strategies are to prove formulas about program properties that may indicate presence of errors. Andrey Belevantsev presented static analysis techniques for programs written in C/C++, Java, Kotlin, Go, and C#. These techniques includes intraprocedural analysis based on value ids, interprocedural summary based analysis, static symbolic execution, and some auxiliary algorithms [2]. This approach had been implemented in the Svace static analyzer family. Nikolay Shilov presented alias calculus for a model procedural programming language MoRe that has addressable memory. This calculus has been implemented in an aliasing analysis prototype tool for MoRe [27]. Presented calculus allows finding memory leak bugs in a number of short snippets of MoRe code.

3 Summary and Concluding Remarks A comprehensive study of research seminars history and experience during and after the pandemic caused by COVID-19 outbreak is a topic for more systematic and longterm sociological research, definitely beyond this ruSTEP experience report. We are aware just about a single publication on the topic [6]. At the same time, community concern about computer conferencing tools for collaborative online seminars may be tracked back to XX century (e.g. [9]). In contrast, issues related to education is already a hot topic (e.g. [19]). If to speak about ruSTEP experience, it is possible to say that the seminar has considered different approaches and views to different problems that are important in practice. The seminar, being interinstitutional and connecting people from different physical locations, was planned for a hybrid mode from the start. As such outside

380

N. V. Shilov et al.

events played a little role in the technological solutions employed in the organization of the seminar. However, the recent COVID-19 pandemic still had some effect. In particular, many sessions had been moved entirely to the online format, unifying the experience for all participants. Additionally, the general perception of remote participation has changed. The use of tools like Skype and Zoom has been normalised throughout the pandemic and has lost some of the negative connotations typically associated with remote participation. That said, we believe the effect of the pandemic on the seminar has been minor. Seminar community is looking forward to year 2023 (and beyond) with a hope to resolve problems that ruSTEP met during its first 30 months of history: • Loss of international audience after launch of the special military operation in Ukraine • Drop in number of regular participants after return to offline teaching world-wide • Few hybrid meetings (only in Innopolis) with offline and online presentation and attendance • Lack of advance agenda/meetings planning and a weak engagement with invited presenters after talks • Irregular publication and promotion of seminar activities in appropriate venues (newsletters, workshops, etc.) • Use of email-based communication platform for post-seminar discussions • Poor (or even absence) of integration with educational programs in Innopolis and Novosibirsk State Universities. We hope to resolve all these problems and warmly invite new contributors (speakers and presenters) as well regular participants (from researchers, faculty, and students) to join us! Acknowledgements On behalf of Steering Committee and all participant we would like to thanks. • Innopolis University for providing classes for ruSTEP offline meetings as well as for communication infrastructure that makes online meetings possible. • A.P. Ershov Institute of Informatics Systems for hosting of PDF-archive of the seminar and use of Institute’ YouTube-channel for video-archive (on special channel https://clck.ru/33VMc5).

References 1. Anureev, I.S.: Operational ontological approach to formal programming language specification. Prog. Comput. Softw. 35(1), 35–42 (2009) 2. Belevantsev, A., Avetisyan, A.: Multi-level static analysis for finding error patterns and defects in source code. In: Petrenko, A., Voronkov, A. (eds.) PSI 2017, LNCS, vol. 10742. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-74313-4_3 3. Bodin, E.V., Garanina, N.O., Shilov, N.V.: Mars Robot puzzle (a multiagent approach to the Dijkstra problem). Model. Anal. Inform. Syst. 18(2), 113–128 (2011) (In Russian) 4. Burmyakov, A., Nikoli´c, B.: An exact comparison of global, partitioned, and semi-partitioned fixed-priority real-time multiprocessor schedulers. J. Syst. Arch. 121, 102313 (2021)

Running Regular Research Seminar Online

381

5. Carrasquel Gamez, J.C., Lomazova, I.A., Rivkin, A.: Modeling trading systems using Petri net extensions. In: Proceedings of the International Workshop on Petri Nets and Software Engineering Co-located with 41st International Conference on Application and Theory of Petri Nets and Concurrency (PETRI NETS 2020), vol. 2651, pp. 118–137. CEUR Workshop Proceedings (2020) 6. Davies, A., Seaton, A., Tonooka, C., White, J.: Covid-19, online workshops, and the future of intellectual exchange. Rethinking Hist. 25(2), 224–241 (2021). https://doi.org/10.1080/ 13642529.2021.1934290 7. Denney, E., Fischer, B.: Explaining verification conditions. In: Meseguer, J., Ro¸su, G. (eds.) AMAST 2008, LNCS, vol. 5140, pp. 145–159. Springer, Berlin, Heidelberg (2008). https:// doi.org/10.1007/978-3-540-79980-1_12 8. Garanina, N., Anureev, I., Sidorova, E., Koznov, D., Zyubin, V., Gorlatch, S.: An ontology-based approach to support formal verification of concurrent systems. In: Sekerinski, E., Moreira, N., Oliveira, J.N., Ratiu, D., Guidotti, R., Farrell, M., Luckcuck, M., Marmsoler, D., Campos, J., Astarte, T., Gonnord, L., Cerone, A., Couto, L., Dongol, B., Kutrib, M., Monteiro, P., Delmas, P. (eds.) FM 2019, LNCS, vol. 12232, pp. 114–130. Springer, Cham (2020). https://doi.org/ 10.1007/978-3-030-54994-7_9 9. Juell, P., Brekke, D., Vetter, R., Wasson, J.: Evaluation of computer conferencing tools for conducting collaborative seminars on the internet. Int. J. Educ. Telecommun. 2(4), 233–248 (1996) 10. Kogtenkov, A.V.: Mechanically proved practical local null safety. Proc. Inst. Syst. Prog. RAS 28(5), 27–54 (2016) 11. Kondratyev, D., Promsky, A.: Correctness of proof strategy for the sisal program verification. In: Proceedings of the 2019 International Multi-Conference on Engineering, Computer and Information Sciences (SIBIRCON-2019), pp. 641–646. IEEE (2019) 12. Kondratyev, D.A., Promsky, A.V.: Developing a self-applicable verification system. Theory and practice. Autom. Control Comput. Sci. 49(7), 445–452 (2015) 13. Kondratyev, D., Promsky, A.: Proof strategy for automated sisal program verification. In: Mazzara, M., Bruel, J.-M., Meyer, B., Petrenko, A. (eds.) TOOLS 2019, LNCS, vol. 11771, pp. 113–120. Springer, Cham (2019). https://doi.org/10.1007/978-3-030-29852-4_9 14. Kondratyev, D.A., Nepomniaschy, V.A.: Automation of C program deductive verification without using loop invariants. Prog. Comput. Softw. 48(5), 331–346 (2022) 15. Kudasov, N., Sim, V.: Formalizing ϕ-calculus: a purely object-oriented calculus of decorated objects (2022). CoRR abs/2204.07454, https://arxiv.org/abs/2204.07454 16. Kudasov, N.: E-unification for second-order abstract syntax (2023). CoRR abs/2302.05815, https://arxiv.org/abs/2302.05815 17. Kukharenko, V.A., Ziborov, K.V., Sadykov, R.F., Naumchev, A.V., Rezin, R.M., MerkinJanson, L.A.: InnoChain: a distributed ledger for industry with formal verification on all implementation levels. Model. Anal. Inform. Syst. 27(4), 454–471 (2020) (In Russian) 18. Maryasov, I.V., Nepomniaschy, V.A., Promsky, A.V., Kondratyev, D.A.: Automatic C program verification based on mixed axiomatic semantics. Autom. Control Comput. Sci. 48(7), 407–414 (2014) 19. Mazzara, M., Succi, G., Tormasov, A.: Online and blended education: after COVID-19. In: Innopolis University—From Zero to Hero. Springer, Cham (2022). https://doi.org/10.1007/ 978-3-030-98599-8_10 20. Moore, J.S.: Milestones from the pure lisp theorem prover to ACL2. Formal Aspects Comput. 31(6), 699–732 (2019) 21. Moriconi, M., Schwartz, R.L.: Automatic construction of verification condition generators from Hoare logics. In: Even, S., Kariv, O. (eds.) ICALP 1981, LNCS, vol. 115, pp. 363–377. Springer, Heidelberg (1981). https://doi.org/10.1007/3-540-10843-2_30 22. Naumchev, A: Seamless object-oriented requirements. In: Proceedings of the 2019 International Multi-Conference on Engineering, Computer and Information Sciences (SIBIRCON-2019), pp. 743–748. IEEE (2019)

382

N. V. Shilov et al.

23. Nepomniaschy, V.A.: Symbolic method of verification of definite iterations over altered data structures. Prog. Comput. Softw. 31(1), 1–9 (2005) 24. Pitts, A.M., Matthiesen, J., Derikx, J.: A dependent type theory with abstractable names. Electron. Notes Theor. Comput. Sci. 312(C), 19–50 (2015) 25. Romanov, V., Ivanov, V.: Prediction of types in python with pre-trained graph neural networks. In: Proceedings of the 2022 Ivannikov Memorial Workshop (IVMEM-2022), pp. 54–60. IEEE (2022) 26. Rosu, G.: Matching logic. Logical Methods Comput. Sci. 13(4), 1–61 (2017) 27. Shilov, N., Satekbayeva, A., Vorontsov, A.: Alias calculus for a simple imperative language with decidable pointer arithmetic. Bull. Novosibirsk Comput. Center. Ser. Comput. Sci. (37), 131–147 (2014) 28. Slind, K., Norrish, M.: A brief overview of HOL4. In: Mohamed, O.A., Muñoz, C., Tahar, S. (eds.) TPHOLs 2008, LNCS, vol. 5170, pp. 28–32. Springer, Berlin, Heidelberg (2008). https://doi.org/10.1007/978-3-540-71067-7_6 29. Staroletov, S.: Towards modeling and verification of eurobalise telegram encoding algorithm. Transp. Res. Procedia 61, 447–454 (2022) 30. Ushakova, M.S., Legalov, A.I.: Verification of programs with mutual recursion in Pifagor language. Autom. Control Comput. Sci. 52(7), 850–866 (2018) 31. Zyubin, V., Anureev, I., Garanina, N., Staroletov, S., Rozov, A., Liakh, T.: Event-driven temporal logic pattern for control software requirements specification. In: Hojjat, H., Massink, M. (eds.) FSEN 2021, LNCS, vol 12818, pp. 92–107. Springer, Cham (2021). https://doi.org/10. 1007/978-3-030-89247-0_7

Innopolis University: An Agile and Resilient Academic Institution Navigating the Rocky Waters of the COVID-19 Pandemic Yuliya Krasylnykova, Iouri Kotorov , Jaroslav Demel, Manuel Mazzara , and Evgeny Bobrov

Abstract Fierce winds produced by the Fourth Industrial Revolution, socioeconomic upheavals, geopolitical strife and other black swan events can lead to dangerous water conditions threatening the business and operational continuity of organisations. The unprecedented disruptions caused by COVID-19 have significantly impacted all industries worldwide, and the education sector is no exception. The pandemic forced university leaders to find ways to stay afloat on the fly and make swift decisions to ensure an uninterrupted flow of education. One such decision was the adoption of digital technologies. Whilst many higher education institutions had to revamp their strategies and redesign their business models and processes, Innopolis University continued to operate with minimal interruption. It switched to distance learning with minor problems, and its business model remained intact. Relying on its institutional values and emulating the moves of those in agile and resilient organisations, Innopolis University was able to stand firm in the storm and become better because of it. Unequivocally, universities are not all in the same boat, but we are all in the same sea. This paper shines a light on the University’s “sea-going” experience and strategies that helped it remain buoyant, which can be valuable to any institution of tertiary education worldwide.

Y. Krasylnykova (B) · I. Kotorov Karelia University of Applied Sciences, Joensuu, Finland e-mail: [email protected] I. Kotorov Université Paul Sabatier, IRIT, Toulouse, France J. Demel Technical University of Liberec, Liberec, Czech Republic M. Mazzara · E. Bobrov Innopolis University, Innopolis, Republic of Tatarstan, Russia © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_36

383

384

Y. Krasylnykova et al.

1 The Pilot Card of Innopolis University Innopolis University (IU) [1], established on 10th December 2012, is a one-of-a-kind English-speaking university of Information Technology (IT) located in Innopolis City, the Republic of Tatarstan, Russia. The city of self-driving cars, situated just a few kilometres from the republic’s capital Kazan, was inaugurated on 9th June 2015 and is home to the Innopolis Special Economic Zone, Information and Communications Technology (ICT) businesses and ambitious start-ups. Ab incunabulis, Innopolis University has been committed to the highest international academic standards aspiring to become the major Russian IT hub [2]. Being an English-taught university, it offers its students the opportunity to study Computer Science and Engineering in English, providing a twofold benefit (one of the world’s most in-demand professions and true fluency in the English language) as a result [3]. The University’s curriculum is benchmarked against international standards to provide world-class IT education. Moreover, the University articulates industry and university collaboration as its strategic imperative, thus forming a loop where the University acts as a supplier of the highly skilled workforce and the industry as a key player in informing the choices related to educational and research programmes of the University [4–6]. It is also worth noting that the Revenue Streams building block of the University’s business model excludes international students as an income stream. Internationalisation has also been considered a strategic priority from the outset of the University [7]. The University has built an extensive international network of collaboration partners with which it has executed bilateral and multilateral agreements for international student and staff mobility, joint PhD supervision, internships (both online and offline) and joint ventures, to name just a few [8]. In addition, IU regularly organises and hosts a range of academic activities, such as international summer schools, boot camps, competitions, hackathons and conferences [1]. It also invites leading experts to give a talk on pressing issues and international talents who have made names for themselves in their respective areas of expertise as visiting professors to instruct students or perform research [8]. Entrance into rankings, much coveted by many higher education institutions (HEIs), is extremely challenging for any university, particularly for a young one [9]. Nonetheless, IU confidently takes its first faltering steps to gain standing in the rankings. In 2018, the University entered Round University Ranking [10] and scored highly in Internationalization. It received the maximum possible points for “Share of international co-authored papers”, becoming the best among Russian universities, and for “International Diversity”, coming the second best, with only M. V. Lomonosov Moscow State University ahead. Innopolis University is also proud of its core values [1]. Innovation, agility and resiliency, among others, are the guiding principles that anchor almost all work done in the University, inform all its decision, and exemplify the University’s culture [8]. These attributes are deemed to be vital for organisational efficiency, productivity, effectiveness and performance. At the same time, rough waters can occur at any time.

Innopolis University: An Agile and Resilient Academic Institution …

385

Global events are like sea beasts that can turn even the calmest waters treacherous. Here again, innovation, agility and resiliency help smooth the rough waters of the Digital Age and successfully navigate through dangerous sailing conditions created by unpredictable weather of the internal and external environment [11].

2 Innopolis University Sailing into the COVID-19 Storm on the Agile Resiliency The coronavirus disease (COVID-19) first emerged in Wuhan, China, and in a matter of days, declared by the World Health Organisation (WHO) as a “Public Health Emergency of International Concern” [12] caused rough waters on the otherwise beautiful sunny day of January 2020. Within a few weeks, this deadly virus spread all over the world to almost 8 billion people [13], sowing unbridled chaos and fear and making all of us contend with a “new normal”: lockdowns, quarantines, unprecedented safety measures, social distancing, working from home, online shopping, remote teaching, travel bans and the face mask as an emblem of the year [14]. To check the spread of the airborne SARS-CoV-2 virus, many states imposed lockdowns and mandated the shutdown of all non-essential services [14], compelling education entities to close their doors and send their students home [15]. University leaders had to make swift decisions and take decisive actions to maintain academic continuity. They were prompted to turn to ICTs for quick fixes, and Emergency Remote Teaching (ERT) came to their rescue [16]. Oddly enough, the COVID-19 pandemic day out “stress-tested” the agility and resilience of HEIs all over the world, ruthlessly separating winners and losers based on these two qualities [17]. Although agility and resiliency are largely synonymous, the two are not the same [18]. Agility is a term that was coined in the Agile Manifesto [19] and was initially used to describe software development management. It refers to “the capacity for moving quickly, flexibly, and decisively in anticipating, initiating, and taking advantage of opportunities and avoiding any negative consequences of change” [20]. In turn, organisational resilience is defined as “the capacity for resisting, absorbing, and responding, even reinventing if required, in response to fast and/ or disruptive change that cannot be avoided” [20]. Both agility and resiliency are essential for turning adversity into an advantage, as “over-attention to one or the other capability can be disastrous” [20]. By now, it should be clear why only resilient HEIs with agile processes were less likely to capsize and founder in the COVID-19 storm and why they were the only ones to hold a clear advantage over others and, in the upshot, be acknowledged as “true winners” [21]. This paper illuminates how Innopolis University, firmly attached to its core values, managed to tackle the violent waters of the pandemic, won its battle against the COVID-19 monster and emerged from this crisis better than ever before.

386

Y. Krasylnykova et al.

2.1 COVID-19 the Kraken or 2020 Challenge Accepted: Inventory of Life-Saving Appliances and Arrangements For Innopolis University, news of the pandemic was not a reason to panic. Instead, it was a wake-up call to be up to the challenge and create an effective action plan that set the conditions for a powerful short-term response to the COVID-19 crisis and fostered long-term resilience throughout the institution. The word “crisis”, a Latinized form of the Greek “krisis”, means “decision” and refers to a vitally important moment to make decisions about what is essential and what is not. In charting the course ahead, our supervisory priorities were: (1) people, (2) operational and academic continuity and (3) the future state of the University. The actions taken by Innopolis University in these three dimensions were as follows: Priority no.1: People. With COVID-19 being, first and foremost, a public health emergency, Innopolis University prioritised its staff and students’ health and safety first. In line with government regulations, the University adopted and strictly followed all preventive measures like masks, temperature checks, social distancing, airing and regular cleaning. This also included switching from fully in-person instruction to ERT. Priority no.2: Operational and academic continuity. Here, the key was developing a system that could cater for the dynamic environment of the pandemic and allow for quick pivots and efficient decision-making. Focus was on speed, efficiency, flexibility and practicality. To maintain learning continuity, the University adopted distance learning solutions. As the proper technology conditions the effectiveness of remote teaching strategies, the University reassessed its ICT infrastructure, i.e. hardware, software, the Internet and connectivity, to minimise instances of digital sabotage and ensure a smooth interaction between its students and teaching staff. Priority no.3: The future state of the University. Any crisis brings both difficulties and opportunities. Rather than viewing cascading disruptions caused by the pandemic as glass-half-empty situations, we intended to learn from this experience and use that learning to create a cascading success in the future and thus fill the glass to the brim. Any success is the result of attitude, decisions and actions. We aimed for a truly sustainable future. We decided what must not change (our shared values and caring about each other) and what we could and should change, doubling down on what was working and jettisoning everything that was not. We were determined to win and make lasting improvements to business as usual rather than simply build back better.

2.2 Suddenly Overboard: Repackaging Pedagogy The initial wave of lockdowns in mid-March 2020 turned our daily lives upside down. Nearly overnight, we were left tête-à-tête with the “new normal”. In education, this “new normal” was quite an affecting spectacle: campuses sans students, students

Innopolis University: An Agile and Resilient Academic Institution …

387

stranded at home, wherever home was and an instantaneous radical change in how we educated them. But even with COVID-19 as an omnipresent backdrop, educational institutions had to continue providing adequate instruction to students [14]. Abrupt mandatory campus closures and suspension of all in-person activities led university leaders to find viable “first-aid” solutions, and online learning—in the form of ERT— threw them a lifeline. Online education, both synchronous and asynchronous, is certainly not a recent invention. It has existed since long and is nothing new [22]. At least once, each of us listened to a TED podcast, completed an online course, used YouTube to search for videos to supplement classes, and even flipped the classroom to promote active learning. Nonetheless, not all of us were prepared for such a sudden shift to a virtually new way of education. It is worth pointing out that online and distance learning are not necessarily the same thing. Reference [23] outlined the difference between these two modes of instructional delivery and provided suggestions on how to implement temporary (i.e. ERT) and permanent (i.e. online learning or eLearning) solutions based on the strengths and weaknesses of an academic institution. In its turn, [24] highlighted three success factors for eLearning. They are: “technology, the instructor, and the previous use of technology from the perspective of students, with an emphasis on the instructor as the central role for online education success” [25]. Simply put, eLearning is not about mirroring a physical classroom online, or teaching-by-telling with a webcam plonked [26]. It requires considerable time resource, substantial investment and completely different capabilities and infrastructure for inclusive design and implementation [27]. Let’s admit it, short-term ERT solutions adopted in response to COVID-19 barely resembled the sophisticated and deliberate eLearning strategies covered in the research literature [25]. But even though our teaching from kitchens and backyards and learning with slippers and pyjamas on was a misery at times, online distance learning proved to be a handy tool for imparting knowledge and a powerful instructional response considering the pandemic’s time. Welcome to Zoomlandia, a safe haven in the turbulent seas of COVID-19 For Innopolis University, predominantly inhabited by digital citizens, the switch to ERT did not turn out to be particularly hard. On the whole, it was well-managed and well-received. Being technologically savvy enough, we quickly assessed major online teaching and learning models based on their functionalities. As instruction delivery modes, we adopted synchronous teacher-directed live streaming and asynchronous video-based flipped learning as they were better positioned to meet our teaching and students’ learning needs and could be carried out through various devices. Live-streamed classes were conducted via several video conferencing apps, for instance, Zoom, Microsoft Teams, Google Meet and Lark. As for video-based flipped learning, we either directed students to video resources readily-available online or created video clips ourselves using a number of digital tools, e.g. YouTube, Edpuzzle and Thinglink.

388

Y. Krasylnykova et al.

Although almost nothing can surpass relational teaching, it is not to say that online learning has no value at all [28]. In our case, it worked perfectly well with more introverted students who were likelier to give presentations and participate in class or group discussions when they were online. Also, the fun fact is that virtually all students found video-recorded lectures (as part of flipped learning) even more valuable compared to live lectures. And it is no wonder. Recorded lecture material allowed students to (re)watch the recorded lectures at their convenience and at their own speeds (using video-accelerating technologies) without distorting the pitch of the instructor’s voice. Putting testing to the test: a creative overhaul of traditional exams Formative and summative assessments are integral to the learning journey and key tools instructors use to determine whether a student has achieved learning outcomes [29]. However, COVID-19 revealed that traditional assessment practices were losing their validity in ERT and had to be updated alongside teaching and learning provision [16]. At Innopolis University, the main issue was the delivery of formal summative assessments, with remote proctoring, security and privacy as primary concerns. Despite being well aware of the potentials offered by AI-enabled assessment systems, de facto, it was neither a wise nor rapid solution we required to keep up with the pace of the pandemic. We were joshing back then that knowing what not to do with technology is a key digital competence. We decided to temporarily throw summative assessments aside and instead continue looking for more flexible, practical and effective ways of testing and grading students’ knowledge. Surprisingly, we found that smart and creative application of alternative assessment arrangements, exempli gratia, group projects, presentations, performance tasks, reflective videos, peer and self-assessments, etc., could be just as valid as pen-and-paper summative tests. This is not to mention that such substitution helped our students be less stressed, more curious and overall more involved in the academic enquiry process—something that, for us, given the complex realities of COVID-19 lockdowns, was more precious than a multitude of rubies.

2.3 Beyond Access: Leveraging Open Educational Resources to Support Students and Faculty in Emergency Distance Learning With the ongoing spread of the disease, libraries were forced to close their on-site services, thus limiting students’ access to learning resources [30]. Students unable to access library collections had to be content with the material they could find online [31]. The library staff of Innopolis University tried to be as responsive to the needs of its community as possible. They revised their practices and innovated their

Innopolis University: An Agile and Resilient Academic Institution …

389

services, rapidly adapting to the dynamics of the pandemic. In collaboration with programme managers and individual faculty members, the library staff helped to identify the availability of digital equivalents of required textbooks, prioritising them for purchase. To support learning during the COVID-19 crisis, the University library undertook advocacy of Open Educational Resources (OERs) as a more agile solution. Creative Commons (CC) defines OERs as “teaching, learning, and research materials that are either (a) in the public domain or (b) licensed in a manner that provides everyone with free and perpetual permission to engage in the 5R activities– retaining, remixing, revising, reusing and redistributing the resources” [32]. They include “full courses/ programmes, course materials, modules, student guides, teaching notes, textbooks, research articles, videos, assessment tools and instruments, interactive materials such as simulations and role plays, databases, software, apps (including mobile apps) and any other educationally useful materials” [33]. Simply put, OERs encompass the myriad of educational materials that are “openly available for use without an accompanying need to pay royalties or licence fees” [34]. Although the University’s professors preferred to stick to the coursebooks adopted pre-pandemic, we saw such times of flux as an invitation for experimentation. We embraced OERs as a conduit for creating more active, engaging distance learning experiences. Supplementing formal online classes with OERs helped us enrich learning and foster student involvement in the digital environment. They also served as a great source of inspiration for designing formative evaluation activities for students. In sum and substance, the effective use of OERs helped us build greater resilience within the University’s learning ecosystem. In addition, OERs, particularly “How-to” videos, helped the University promote its students’ emotional wellness and nurture their resilience during the COVID-19 lockdowns.

3 Conclusion Without bias or prejudice, 2020 was a year of tremendous struggle, hardship and drama as the public health crisis that rocked the world collided with daily life. Nonetheless, despite being the talk of the whole world and a synonym for all that was wrong, the pandemic turned out to be a blessing in disguise for the education sector. Like a tsunami that rushes away from the shore, leaving an uncovered stretch of land, COVID-19 exposed all vulnerabilities and shortcomings of what happens with technology-based distance learning when it is ushered at a massive scale. All this flotsam, reeking and being impossible to ignore, catalysed blue-sky thinking, pivoting and innovation, engendering opportunities for educational establishments to break from past inertia, rethink academic practices, leverage technology and digitalisation and reinvent [35]. Ironically, merciless COVID-19 also subjected all educational institutions to an extreme trial of organisational robustness, drawing a line between winners and losers

390

Y. Krasylnykova et al.

with just two attributes, agility and resiliency, as a key differentiator [36]. And it is not difficult to guess why. Agility is the ability to “act fast, flexibly and decisively” [37]. Resiliency is the capability to “weather adversity, emerge stronger and thrive through crises” [38]. That’s why only those institutions that respond to changes at the speed they require and take advantage of turbulent times, i.e. agile and resilient, could stand a chance of winning [20]. Someone once said: “Winners focus on winning. Losers focus on winners.” In business, however, winning, more often than not, starts with the business, not its workforce [39]. And that means culture, which is, according to Peter Drucker, often saluted as “the Albert Einstein of Management”, “the most critical element in building a winning team” [40]. At Innopolis University, our culture fueled our success. We were not willing to squander the plethora of opportunities offered by the COVID-19 pandemic and used every challenge it threw at us as a chance to celebrate resiliency, creativity and innovation. We were determined to win, and we won. Whether we like it or not, the harsh fact of life we must all accept is that the winds of change are blowing straight into the face of time, radically reshaping the global business landscape. Crises are inevitable. Change is the only constant [41]. Although it is impossible to control the weather, academic settings aren’t helpless. By cultivating a culture of resiliency, educational institutions may minimise the chances of breaking apart when sailing rough, learn to fancy big waves and embrace even the choppiest, rockiest cruises, as only rough seas make the best sailors.

References 1. Innopolis University. https://innopolis.university/en/ 2. Yuloskov, A., Bahrami, M.R., Mazzara, M., Kotorov, I.: Smart cities in Russia: current situation and insights for future development. Future Internet. 13, 252 (2021). https://doi.org/10.3390/ fi13100252 3. Aslam, H., Naumcheva, M., Zhdanov, P., Kotorov, I., Mazzara, M., Akhmetgaraeva, E., Valiev, R., Krasylnykova, Y.: Perception of the Internationalization Process by the University Employees: The Case Study of Innopolis University (2022) 4. Kotorov, I., Krasylnykova, Y., Zhdanov, P., Mazzara, M., Aslam, H., Akhmetgaraeva, E., Naumcheva, M., Brown, J.A.: Institutional commitment and leadership as prerequisites for successful comprehensive internationalization. In: Succi, G., Ciancarini, P., Kruglov, A. (eds.) Frontiers in Software Engineering, pp. 1–11. Springer International Publishing, Cham (2021) 5. Blaschke, P., Demel, J., Kotorov, I.: Innovation Performance of Small, Medium-Sized, and Large Enterprises in Czechia and Finland. Technical University of Liberec, Liberec, Czech Republic (2021) 6. Blaschke, P., Demel, J., Kotorov, I.: Innovation performance of Czech and Finnish manufacturing enterprises and their position in the EU. ACC J. 27, 7–21 (2021). https://doi.org/10. 15240/tul/004/2021-2-001 7. Kotorov, I., Krasylnykova, Y., Zhdanov, P., Mazzara, M.: Finding the right understanding: twenty-first century university, globalization and internationalization. In: Bruel, J.-M., Capozucca, A., Mazzara, M., Meyer, B., Naumchev, A., Sadovykh, A. (eds.) Frontiers in Software Engineering Education, pp. 341–353. Springer International Publishing, Cham (2020) 8. Kotorov, I., Krasylnykova, Y., Zhdanov, P., Mazzara, M.: Internationalization strategy of Innopolis University. In: Bruel, J.-M., Capozucca, A., Mazzara, M., Meyer, B., Naumchev, A.,

Innopolis University: An Agile and Resilient Academic Institution …

9. 10. 11. 12. 13. 14. 15.

16.

17. 18.

19. 20. 21.

22. 23. 24. 25.

26. 27.

28.

29.

391

Sadovykh, A. (eds.) Frontiers in Software Engineering Education, pp. 327–340. Springer International Publishing, Cham (2020) Pagell, R.: Bibliometrics and university research rankings demystified for librarians. In: Library and Information Sciences: Trends and Research. pp. 137–160 (2014) Round University Ranking. https://roundranking.com/ Abad-Segura, E., Gonz, M.-D.: Sustainable management of digital transformation in higher education: global research trends 24 (2020) Archived: WHO Timeline - COVID-19. https://www.who.int/news/item/27-04-2020-who-tim eline---covid-19 World Population Clock: 8 Billion People (LIVE, 2023) – Worldometer. https://www.worldo meters.info/world-population/ Gruenwald, H.: Covid19 New Normal Economy (2020) Leal Filho, W., Price, E., Wall, T., Shiel, C., Azeiteiro, U.M., Mifsud, M., Brandli, L., Farinha, C.S., Caeiro, S., Salvia, A.L., Vasconcelos, C.R., de Sousa, L.O., Pace, P., Doni, F., Veiga Avila, L., Fritzen, B., LeVasseur, T.J.: COVID-19: the impact of a global crisis on sustainable development teaching. Environ. Dev. Sustain. (2021). https://doi.org/10.1007/s10668-020-011 07-z Jili, N.N., Ede, C.I., Masuku, M.M.: Emergency remote teaching in higher education during Covid-19: challenges and opportunities. Int. J. Higher Educ. 10, 1 (2021). https://doi.org/10. 5430/ijhe.v10n5p1 Sá, M.J., Serpa, S.: The COVID-19 pandemic as an opportunity to foster the sustainable development of teaching in higher education 16 (2020) Mangalaraj, G., Nerur, S., Dwivedi, R.: Digital transformation for agility and resilience: an exploratory study. J. Comput. Inf. Syst. 63, 1–13 (2022). https://doi.org/10.1080/08874417. 2021.2015726 Manifesto for Agile Software Development. http://agilemanifesto.org/ McCann, J., Selsky, J.W.: Mastering Turbulence: The Essential Capabilities of Agile and Resilient Individuals, Teams and Organizations. Jossey-Bass, San Franciso (2012) McCormack, T.J., Lemoine, P.A., Waller, R.E., Richardson, M.D.: Global higher education: examining response to the COVID-19 pandemic using agility and adaptability. J. Educ. Dev. 5, 10 (2021). https://doi.org/10.20849/jed.v5i1.848 Yamagata-Lynch, L.C.: Blending online asynchronous and synchronous learning. Int. Rev. Res. Open Distrib. Learn. 15 (2014). https://doi.org/10.19173/irrodl.v15i2.1778 Hodges, C.B., Moore, S., Lockee, B.B., Trust, T., Bond, M.A.: The Difference Between Emergency Remote Teaching and Online Learning (2020) Volery, T., Lord, D.: Critical success factors in online education. Int. J. Educ. Manag. 14, 216–223 (2000). https://doi.org/10.1108/09513540010344731 Mohapatra, B.P., Nanda, S.S., Hiremath, C.V., Halagatti, M., Das, S.C., Das, A.: Factors influencing the success of online education during COVID-19: a case analysis of Odisha. J. Risk Financ. Manag. 16, 141 (2023). https://doi.org/10.3390/jrfm16030141 Zhang, D., Zhao, J.L., Zhou, L., Nunamaker, J.F.: Can e-learning replace classroom learning? Commun. ACM. 47, 75–79 (2004). https://doi.org/10.1145/986213.986216 Kloos, C.D., Alario-Hoyos, C., Morales, M., Rocael, H.R., Jerez, Ó., Pérez-Sanagustín, M., Kotorov, I., Fernández, S.A.R., Oliva-Córdova, L.M., Solarte, M., Jaramillo, D., Teixeira, A.M., López, A.H.G.: PROF-XXI: teaching and learning centers to support the 21st century professor. In: 2021 World Engineering Education Forum/Global Engineering Deans Council (WEEF/GEDC). pp. 447–454 (2021) Alqahtani, A.Y., Rajkhan, A.A.: E-learning critical success factors during the COVID-19 pandemic: a comprehensive analysis of e-learning managerial perspectives. Educ. Sci. 16 (2020) ˇ Dedze, I.: Barriers and drivers of innovation in higher education: Lašáková, A., Bajzíková, L, case study-based evidence across ten European universities. Int. J. Educ. Dev. 55, 69–79 (2017). https://doi.org/10.1016/j.ijedudev.2017.06.002

392

Y. Krasylnykova et al.

30. Fabriz, S., Mendzheritskaya, J., Stehle, S.: Impact of synchronous and asynchronous settings of online teaching and learning in higher education on students’ learning experience during COVID-19. Front. Psychol. 12 (2021) 31. Crawford, J., Butler-Henderson, K., Rudolph, J., Malkawi, B., Glowatz, M., Burton, R., Magni, P.A., Lam, S.: COVID-19: 20 countries’ higher education intra-period digital pedagogy responses. J. Appl. Learn. Teach. 3, 09–28 (2020). https://doi.org/10.37074/jalt.2020. 3.1.7 32. Open Education. https://creativecommons.org/about/program-areas/education-oer/ 33. UNESCO: Guidelines for open educational resources (OER) in higher education (2015) 34. Thomas, M.: A basic guide to open educational resources. Br. J. Edu. Technol. 43, E171–E172 (2012). https://doi.org/10.1111/j.1467-8535.2012.01378.x 35. Kotorov, I., Krasylnykova, Y., Demel, J., Blaschke, P.: The Effect of the Covid-19 Pandemic on Economic Growth and R&D Spending in the EU Countries (2023) 36. Mete, J.K., Das, R., Chowdhury, A.: Post-covid challenges and opportunities for higher education. J. Higher Educ. Theo. Pract. 22 (2022). https://doi.org/10.33423/jhetp.v22i9.5371 37. Trotter, H., Huang, C.-W., Czerniewicz, L.: Seeking equity, agility, and sustainability in the provision of emergency remote teaching during the COVID-19 pandemic: a center for teaching and learning takes an expanded role. Higher Learn. Res. Commun. 12, 1–24 (2022). https:// doi.org/10.5590/10.18870/hlrc.v12i0.1280 38. Trump, B.D., Linkov, I.: Risk and resilience in the time of the COVID-19 crisis. Environ Syst. Decis. 40, 171–173 (2020). https://doi.org/10.1007/s10669-020-09781-0 39. Yamaguchi, D., Tezuka, Y., Suzuki, N.: The differences between winners and losers in competition: the relation of cognitive and emotional aspects during a competition to hemodynamic responses. Adapt. Hum. Behav. Physiol. 5, 31–47 (2019). https://doi.org/10.1007/s40750-0180104-5 40. Drucker, P.F.: The Effective Decision. https://hbr.org/1967/01/the-effective-decision (1967) 41. Harari, Y.N.: 21 Lessons for the 21st Century: ‘Truly Mind-Expanding. Ultra-topical’ Guardian. Vintage, London (2019)

Teaching the Future: The Vision of AI/ChatGPT in Education Mohammad Reza Bahrami, Bahareh Bahrami, Farima Behboodi, and Samae Pourrafie

Abstract This article highlights the importance of teaching future technologies in education. With rapid technological advancements, it is crucial for students to be familiar with the skills and knowledge they need to succeed in a future dominated by technology. New technologies like virtual and augmented reality, artificial intelligence and machine learning, are transforming the way of teaching and learning that we get used to. This helps providing new opportunities for personalized and immersive learning experiences. However, the integration of these technologies must be done in an ethical and responsible manner. Considering the potential consequences and implications, and ensuring appropriate safeguards are important features. The article emphasizes on the need to teach future technologies in education to prepare students for the future and teach them the skills they need to thrive in a rapidly changing world.

1 Introduction The traditional educational system for centuries has been the governing mode of education [1, 2]. Here, students sit in a physical classroom and listen to lectures. The assessment is done by taking exams as well as completing the assignments (home works). Typically, the teacher is considered as the main source of earning knowledge from as well as the primary source of instruction. Although the traditional system has been successful in transmitting knowledge, but it has its own limitations and M. R. Bahrami (B) Innopolis University, Innopolis 420500, Russia e-mail: [email protected] B. Bahrami Kerman University of Medical Sciences, 7616913555 Kerman, Iran F. Behboodi Kazan State Medical University, Kazan 420012, Russia S. Pourrafie Kazan Federal University, Kazan 420008, Russia e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_37

393

394

M. R. Bahrami et al.

disadvantages. In this system, rather than critical thinking, creativity, and problemsolving, the focus is on memorization and recall of information (for assessments). The main deficiency of traditional educational system is that it is not capable to keep pace with the rapid advancements in technology and the changing job market [3]. Nowadays, by introducing and development of Artificial Intelligence (AI), AI integrates in educational system [4–7]. This integration rapidly changes the way of teaching and learning that we get used to. AI refers to the development of computer systems that can execute tasks that typically require human intelligence [8]. Tasks like visual perception, speech recognition, decision-making, and language translation are some examples of AI applications. AI systems are designed to learn from data and based on them identify patterns (mining some information) and make predictions (decisions). AI [9] technologies are rapidly improving and increasingly are becoming integrated into our daily lives. The integration of AI in education has the potential to transform teaching and learning. AI-powered learning systems can provide personalized and adaptive learning experiences [10, 11]. According to needs and pace of each individual student the learning method can be tailored with the help of AI. Also, AI can provide prompt feedback to students. Meanwhile, AI can be used as a supporting tool in traditional teaching methods. For example, AI can grade assignments [12] and provide feedback on them, while teachers are more focusing on creating context to ensure the high quality of delivered materials in lectures. AI in education has many potential benefits, however, there are also challenges and concerns that need to be addressed. For example, there are concerns about privacy, ethics, and the potential for AI systems to reinforce existing biases and inequalities [13–17]. It is important for educators and policymakers to carefully consider the implications of AI in education aligned with ethical and educational principles. This is a unique opportunity to rethink and improve the traditional system of education by integrating with AI. This helps students obtaining skills and knowledge they need to thrive in a rapidly changing world. The article emphasizes on the need to teach future technologies in education and educational systems to prepare students for the future and teach them with the skills they need to thrive in a rapidly changing world.

2 AI in Education Artificial Intelligence (AI), is a field of computer science and engineering. AI can perform tasks that normally require human intelligence. This ability (performing tasks) is accomplished by creating algorithms, models and (software) tools. For example, AI can perform tasks like speech recognition, text generating, decision making according to inputs conditions and requirements, image recognition, language translation, etc. [18]. AI tools are designed to simulate human thought processes like learning, reasoning and perception. The idea behind AI is to perform specific tasks more effi-

Teaching the Future: The Vision of AI/ChatGPT in Education

395

ciently and accurately than humans. In general, AI can be classified into two main categories: narrow (weak AI) and general (strong AI) [19]. Narrow AI means that AI system can execute one single task, while general AI system is capable of executing any intellectual task that a human can. Creating systems that are capable of independent thought and decision making is the goal of strong AI. This kind of systems are able to adapt and improve over time (by learning). The advancements in AI have brought about numerous innovations in different industries and fields. It has the potential to change the world in keen ways, helping us to solve complex problems and making our lives easier and more productive. One of its directions is in education. One of the most remarkable ways that AI may be implemented in education is through the AI-powered virtual tutors [10]. These tutors are able to provide individual feedback according to the needs of students. Increase engagement and motivation is another feature. Additionally, AI-powered tools and platforms are being used to analyze student data and provide insights into student learning and performance. This way allows instructors to make more informed decisions about teaching and supporting students. Assessment is another sector that AI can have impact on. AI-powered tools are being used to grade students’ works and provide instant individual feedback. Implementation of these tools significantly reduce the workload of instructors and for sure increase the speed of assessment process.

2.1 AI-Powered Virtual Tutor AI-powered virtual tutor is a AI tool in educational system that was created with the aim of providing individual feedback and support to students. Students’ data (tests scores and past performance) are collected and analysed by this virtual tutor. Then, this data is used to make customized individual lesson plans (syllabus) and exercises. These plans are created according to each student’s needs. One of the main features of AI-powered virtual tutors is providing instant feedback on the students’ works. This may increase the engagement and leads to motivation. These virtual tutors, analysing data, can identify difficult topics/areas for each student and accordingly provide a help (invite to office hours). In addition to providing individual support, AI-powered virtual tutors can be beneficial in increasing the learning process efficiency. This can be accomplished by automating certain tasks like grading and assessment. This can free up time for educators to focus on other important tasks: providing more office hours to work with students or developing new instructional materials. It is important to note, however, that AI-powered virtual tutors should be used as a supplement to human instruction, not as a replacement. Although these virtual tutors are effective at providing feedback and individual support, but they are not able to provide emotional support, empathy, and creativity as human.

396

M. R. Bahrami et al.

By student Solving the assignment

Assignment submission

By AI Understanding the student’s answers and solution

Comparing the student’s answers and solution with the correct ones

Grading

Providing feedback

Fig. 1 Grading scenario by AI tutor

For the rest of this section, we provide a scenario: how AI-powered virtual tutors can grade mathematical assignments [20–22]. First, we may ask: Can AI-powered virtual tutors grade mathematical assignments or tests? The answer is “yes”. AI algorithms can analyze student submissions, evaluate the accuracy of solutions, and provide grades and feedback in real-time. By this, the speed of assessment process will be increased as well as the efficiency of assessment. Here is an example of how an AI-powered virtual tutor can grade a mathematical assignment. The process is shown in Fig. 1. The student solves a mathematical assignment and submits the detailed solutions solving equations. Now, the process of evaluation and grading starts. The AI-powered virtual tutor with the help of image recognition and natural language processing (NLP) understands the student’s answers and solutions [23–25]. Utilizing algorithms virtual tutor then compares the student’s solutions with the correct answers. Then evaluation process starts to check how precise are the solutions. After that, virtual tutor grades the work and provides an individual feedback on the work, particularly, on weaknesses. Finally, the student receives a grade and feedback. This allows him/her to understand where are mistakes and it improves student’s understanding of the material.

Teaching the Future: The Vision of AI/ChatGPT in Education

397

Fig. 2 ChatGPT conversation and prompt

One of the most advantage of grading by the AI-powered virtual tutor is providing immediate feedback and a grade to students. Grading by AI allows students to understand their performance and make improvements and actions more quickly. It is important to note that although AI-powered virtual tools provide accurate and efficient grading, but they should not be considered/used as a substitute for human evaluation. Human instructors should review and verify the grades and feedback provided by these AI tools (virtual tutors) to ensure accuracy and fairness.

3 ChatGPT in Education In this section, since it is about ChatGPT and its role in education, we conduct an experiment to see how ChatGPT can generate text for the scientific purpose. It should be noted that ChatGPT is not the single case of AI text generator, but in this work, we focus on it. In https://chat.openai.com, we wrote the prompt as shown in Fig. 2. The rest of this section is generated by ChatGPT according to our prompt.

3.1 An Introduction to ChatGPT ChatGPT, developed by OpenAI, is a remarkable language model that has been trained on a vast amount of text data. It has the capability to respond to natural language queries with answers that are almost indistinguishable from those generated by humans. This technology is at the forefront of artificial intelligence and has the potential to revolutionize many industries, including education.

3.2 Applications of ChatGPT in Education ChatGPT can be applied in various ways to support education and improve the learning process. Some of the key applications of ChatGPT in education include:

398

M. R. Bahrami et al.

1. Personalized Tutoring: ChatGPT can serve as a virtual tutor, providing students with immediate feedback and assistance. This helps to increase student engagement and motivation and provides support for students who may need extra help. 2. Question Answering: ChatGPT can be utilized as an AI-powered questionanswering system, allowing students to ask questions and receive accurate, concise answers. This helps to increase the efficiency of the learning process and can be particularly helpful for students who are looking for quick answers to specific questions. 3. Content Generation: ChatGPT can generate educational content, such as practice problems, summaries, and study guides. This can help to save time and effort for educators, who can focus on other important tasks, and provide students with access to high-quality educational resources.

3.3 Potential Benefits of Using ChatGPT in Education The use of ChatGPT in education offers numerous benefits, including: 1. Enhanced Engagement: By providing personalized feedback and support, ChatGPT can improve student engagement and motivation. This can help to keep students interested in the learning process and increase their participation. 2. Increased Efficiency: ChatGPT can automate tasks such as grading and assessment, freeing up time for educators to focus on other important tasks. This can help to increase the efficiency of the learning process and improve the quality of education. 3. Personalized Learning: ChatGPT can be used to create customized lesson plans and exercises tailored to the individual needs of each student. This allows for more effective and personalized learning and can help to increase student engagement and motivation. 4. Access to Information: ChatGPT provides students with instant access to information and knowledge, allowing them to expand their understanding and deepen their learning. This helps to increase student engagement and motivation and can lead to improved learning outcomes. In conclusion, ChatGPT represents a major advancement in the field of education and has the potential to revolutionize the way we approach learning. By providing personalized feedback and support, improving the efficiency of the learning process, and increasing access to information, this technology has the power to benefit both students and educators alike.

Teaching the Future: The Vision of AI/ChatGPT in Education

399

4 Challenges and Concerns 4.1 Ethical and Moral Considerations The usage of AI technologies, like ChatGPT, in the educational systems [26] brings a range of ethical and moral considerations that need to be carefully evaluated and addressed. These considerations are the potential for perpetuating biases, the privacy of student data, and the responsibility for ensuring the fair and just use of AI in education. AI technologies like ChatGPT are trained on large amounts of data. If that data contains biases, then the AI tool can produce biased outputs. It is crucial to identify and mitigate these biases to ensure that AI tools are used in a fair way. The privacy of student data are a significant concern. It is imperative to ensure that student data is protected and used only for educational purposes, in compliance with privacy laws and regulations. Also, it is important to consider who is responsible for use of outputs of AI tools.

4.2 Technical Challenges Despite the potential for AI/ChatGPT to greatly magnify education, there are also technical challenges that must be overcome. These challenges are ensuring accuracy in responses, scalability [27] for increasing student queries, and interoperability with existing education systems. AI technologies must provide accurate responses to student queries. To achieve this, the technology must be trained on high-quality, relevant data. This tools should be continuously under monitoring and improvement. AI tools should able to handle a large amount of students’ queries and data. To accomplish this, scalable technology is needed to analyse big data [28, 29]. AI tools must work smoothly with other systems and platforms used in education like Moodle. Therefore, development of contracts [30] are needed for the exchange of data between them.

4.3 Implementation Considerations To successfully integrate AI/ChatGPT into education, there exists several implementation considerations that need to be taken into account. These considerations are presented in Table 1.

400

M. R. Bahrami et al.

Table 1 AI implementation considerations Item Consideration Description 1

Integration with existing systems

2

Training and support

3

Evaluation and improvement

AI tools must integrate with existing education systems, like Moodle [31]. This requires the development of APIs (Application Programming Interface) and other platforms (or new contracts) to facilitate data exchange between AI tools and existing systems Instructors and students must receive training on how to effectively use AI tools in the classroom. This requires the development of training and supportive materials AI tools must be continuously evaluated and improved to meet the needs of instructors and students

5 Future of AI/ChatGPT in Education The usage of AI technologies, like ChatGPT, in educational system is expected to continue its increase in the coming years. Factors like personalized Learning, widespread adoption, accessibility, precise assessment and interdisciplinary applications have big impact on this trend. Personalization [32] is a key aspect of transformation of both the teaching and learning experience by AI/ChatGPT. ChatGPT can tailor educational materials to meet the needs and abilities of individual students. This can be achieved through the development of customized study materials, assignments, and feedback. All of these can be obtained in real-time based on student performance. For example, imagine a student who is struggling with a particular subject. AIpowered virtual tutors and interactive lessons can be used to provide targeted instruction and feedback. This allows the student to focus on his/her weakness and improve his/her understanding of the material. Also, ChatGPT can be used to teach students who who may not have access to traditional classrooms. For example, students with disabilities (because of lack of infrastructure in schools) or students who because of health problem can not attend in person can use AI-powered virtual tutors and interactive lessons. They provide education to these students, regardless of their location or circumstances. AI tools provide opportunities for all learners. The future of education and the educational system is expected to be heavily influenced by the integration of AI and other advanced technologies. The trend towards personalized learning is one of the most promising developments in this regard. By leveraging AI, virtual tutors can create individualized learning experiences that cater to each student’s unique needs and abilities. This can result in improved engagement, motivation, and performance for students.

Teaching the Future: The Vision of AI/ChatGPT in Education

401

Another key trend is the growth of online education [33] integrated with AI virtual tutors. With the increasing use of technology in education, students will have the opportunity to access high-quality education from anywhere in the world, at any time. Particularly for students who may not have the same access to educational resources as others is the best option.

6 Conclusion The integration of AI/ChatGPT in education has the potential to bring about a positive change in the learning experience. Its ability to personalize, streamline, and facilitate education has the potential to prepare the next generation for success in the twentyfirst century. The potential benefits of using ChatGPT in education are a lot. It may improve student engagement and performance, meanwhile reducing the instructors’ workload. AI in education may provide access to education for all learners. Acknowledgements We thank Innopolis University for generously funding this endeavor.

References 1. Mazzara, M., Zhdanov, P., Bahrami, M.R., Aslam, H., Kotorov, I., Imam, M., Salem, H., Brown, J.A., Pletnev, R.: Education after covid-19. In: Smart and Sustainable Technology for Resilient Cities and Communities, pp. 193–207. Springer (2022) 2. Gowda, R.S., Suma, V.: A comparative analysis of traditional education system vs. e-learning. In: 2017 International Conference on Innovative Mechanisms for Industry Applications (ICIMIA), pp. 567–571. IEEE (2017) 3. Naidu., S.: Learning & Teaching with Technology: Principles and Practices. Psychology Press (2003) 4. Chen, L., Chen, P., Lin, Z.: Artificial intelligence in education: a review. IEEE Access 8, 75264–75278 (2020) 5. Roll, I., Wylie, R.: Evolution and revolution in artificial intelligence in education. Int. J. Artif. Intell. Educ. 26, 582–599 (2016) 6. Pedro, F., Subosa, M., Rivas, A., Valverde, P.: Artificial intelligence in education: challenges and opportunities for sustainable development (2019) 7. Holmes, W., Bialik, M., Fadel, C.: Artificial intelligence in education (2020) 8. McCarthy, J., et al.: What is artificial intelligence (2007) 9. Ertel, W.: Introduction to Artificial Intelligence. Springer (2018) 10. Goel, A.: AI-powered learning: making education accessible, affordable, and achievable. arXiv preprint arXiv:2006.01908 (2020) 11. Amer-Yahia, S.: Towards AI-powered data-driven education. Proc. VLDB Endowment 15(12), 3798–3806 (2022) 12. Chassignol, M., Khoroshavin, A., Klimova, A., Bilyatdinova, A.: Artificial intelligence trends in education: a narrative overview. Procedia Comput. Sci. 136, 16–24 (2018) 13. Siau, K., Wang, W.: Artificial intelligence (AI) ethics: ethics of AI and ethical AI. J. Database Manage. (JDM) 31(2), 74–87 (2020) 14. Jobin, A., Ienca, M., Vayena, E.: The global landscape of AI ethics guidelines. Nat. Mach. Intell. 1(9), 389–399 (2019)

402

M. R. Bahrami et al.

15. Hagendorff, T.: The ethics of AI ethics: an evaluation of guidelines. Minds Mach. 30(1), 99–120 (2020) 16. Coeckelbergh, M.: AI Ethics. MIT Press (2020) 17. Shaw, J.: Artificial intelligence and ethics. Harvard Magazine, 30 (2019) 18. Smith, R.G., Eckroth, J.: Building AI applications: yesterday, today, and tomorrow. AI Magazine 38(1), 6–22 (2017) 19. Flowers, J.C.: Strong and weak AI: Deweyan considerations. In: AAAI Spring Symposium: Towards Conscious AI Systems, vol. 2287 (2019) 20. Deo, R.C., Yaseen, Z.M., Al-Ansari, N., Nguyen-Huy, T., Langlands, T.A.M., Galligan, L.: An investigation on engineering mathematics courses: modern artificial intelligence model development for undergraduate student performance prediction. IEEE Access 8, 136697–136724 (2020) 21. Hwang, G.-J., Tu, Y.-F.: Roles and research trends of artificial intelligence in mathematics education: a bibliometric mapping analysis and systematic review. Mathematics 9(6), 584 (2021) 22. Moons, F., Vandervieren, E., Colpaert, J.: Atomic, reusable feedback: a semi-automated solution for assessing handwritten tasks? A crossover experiment with mathematics teachers. Comput. . Open 3, 100086 (2022) 23. Guo, J., He, H., He, T., Leonard Lausen, M., Li, H.L., Shi, X., Wang, C., Xie, J., Zha, S., et al.: GluonCV and GluonNLP: deep learning in computer vision and natural language processing. J. Mach. Learn. Res. 21(1), 845–851 (2020) 24. Bahrami, M.R.: Modeling of Physical Systems Through Bond Graphs. St. Petersburg Polytechnical University, St. Petersburg (2018) 25. Lavanya, P.M., Sasikala, E.: Deep learning techniques on text classification using natural language processing (NLP) in social healthcare network: a comprehensive survey. In: 2021 3rd International Conference on Signal Processing and Communication (ICPSC), pp. 603–609. IEEE (2021) 26. Zhai, X.: ChatGPT user experience: implications for education (2022). Available at SSRN 4312418 27. Weinstock, C.B., Goodenough, J.B.: On system scalability. Technical report, Software Engineering Institute, Carnegie Mellon University, PA, USA (2006) 28. Fan, J., Han, F., Liu, H.: Challenges of big data analysis. Nat. Sci. Rev. 1(2), 293–314 (2014) 29. Sagiroglu, S., Sinanc, D.: Big data: a review. In: 2013 International Conference on Collaboration Technologies and Systems (CTS), pp. 42–47. IEEE (2013) 30. Provost, F., Fawcett, T.: Data science and its relationship to big data and data-driven decision making. Big Data 1(1), 51–59 (2013) 31. Horvat, A., Dobrota, M., Krsmanovic, M., Cudanov, M.: Student perception of Moodle learning management system: a satisfaction and significance analysis. Interact. Learn. Environ. 23(4), 515–527 (2015) 32. Kundu, S.S., Sarkar, D., Jana, P., Kole, D.K.: Personalization in education using recommendation system: an overview. Comput. Intell. Digital Pedagogy, 85–111 (2021) 33. Sun, A., Chen, X.: Online education and its effective practice: a research review. J. Inf. Technol. Educ. 15 (2016)

Thesis Supervision in Computer Science—Challenges and a Gamified Solution Rabab Marouf, Iouri Kotorov, Hamna Aslam, Yuliya Krasylnykova, and Marko Pezer

Abstract During the thesis supervision process, supervisors and supervisees face a number of challenges for the completion of thesis writing in higher education settings. To identify major supervision challenges, survey and interviews have been conducted among current undergraduate and postgraduate students. Computer Science supervisors are also interviewed in parallel to identify their challenges. Based on the comments and feedback from both supervisors and supervises, this study proposes the integration of gamified system to facilitate the process of thesis supervision in computer science.

1 Background In the third decade of the twenty-first century, Higher Education Institutions (HEIs) are affected by rapid transformation processes [1–3]. One of the results of this transformation is the ever-increasing number of students [4, 5]. Still, the number of teachers remains almost unchanged, which puts pressure on students and teachers, especially during the writing of a thesis [6, 7]. Even though students receive a quality education, due to the constant growth in the number of students in groups, many things remain on a more theoretical plane than practical, directly affecting research work at the undergraduate and graduate levels during thesis writing [8, 9]. Students generally have limited independent research experience, and using existing libraries and available literature is also challenging [10]. Thesis writing is an opportunity for students R. Marouf (B) · H. Aslam · M. Pezer Innopolis University, 420500 Innopolis, Russian Federation e-mail: [email protected] I. Kotorov Institut de Recherche en Informatique de Toulouse (IRIT), Université de Toulouse, 31062 Toulouse, France I. Kotorov · Y. Krasylnykova Department of International Business, North Karelia University of Applied Sciences, Karjalankatu 3, 80200 Joensuu, Finland © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_38

403

404

R. Marouf et al.

to gain in-depth knowledge about a research topic and dive deeper into a research question [11]. A thesis bridges the gap between an educational stage and work or further study at the end of a curriculum [11]. In addition to being resource-intensive, thesis supervision is often a critical factor in thesis quality [12]. As the main tasks of any HEIs rest on three pillars: (1) tuition, (2) research and (3) community/ regional/ national service [7, 13]. All three pillars are equally important, but in this paper, we want to focus primarily on the tuition’s final stage, the stage of preparation for writing a thesis and narrow our focus on the main issues associated with the thesis process.

2 State of the Art For many years, the thesis has been regarded as an essential component of the degree programs in HEIs [7, 14]. There are several works covering supervision philosophy and professional practice that span decades [15, 16]. Alternative solutions for completing doctorate theses have been examined, as have advancements in supervision approaches [17]. While many features of postgraduate research supervision are clearly transferrable to undergraduate thesis supervision, there are some notable distinctions [18]. According to [19] undergraduate students have minimal research experience, and transitioning from classroom to independent research has presented challenges and necessitated the need for detailed guidance and supervision. The undergraduate thesis’s significantly shorter duration exacerbates the absence of prior research experience and learning of the corresponding abilities [12, 20]. Among the most significant transformations in student thinking required to transition from guided learning in large group environments to independent learning with supervision [21]. The supervisor-student relationship has a strong influence on the supervision process, conferring to several authors. According to [11, 18, 21] student and supervisor responsibilities are analysed, and their expectations are compared based on who is responsible for what in the supervision process. They found a significant mismatch between supervisors and students regarding who is primarily responsible for which components of this process (for example, selecting the topic, launching meetings, examining progress, discussing methodology, or seeking feedback). To improve results, it is highly recommended that expectations are aligned [15]. Students must meet many criteria to gain the abilities required to reach the objective, for example, they must build in-depth knowledge of a certain topic based on what they have learnt during their studies [14]. They should do a thorough literature review and develop a research question [9, 22]. To work independently, students must establish and adhere to a project plan, which is a commitment to complete each assignment [15]. Finally, students must have excellent communication skills to satisfy supervisors’ expectations [7, 23]. Supervisors also faced several issues [24, 25]. Supervisors should not only be accomplished researchers but also be conversant with research methodology [16]. Furthermore, they should enable higher education students to acquire the necessary abilities for these methodologies [7]. For example, mentoring

Thesis Supervision in Computer Science—Challenges and a Gamified …

405

Table 1 Supervisors’ classification Type

Description

Controller

The thesis and the entire procedure are under the supervision of the supervisor. The supervisor outlines the themes, establishes deadlines, and typically has a definite expectation of the work’s outcome.

Gatekeeper The supervisor collaborates with students who take charge of and organize the thesis writing process. The supervisor has a definite expectation of the outcome and feels personally accountable for it. Supporter

The supervisor specifies the topic of the thesis as well as a defined supervision procedure. Students defined submissions along the process to track the development of the work and receive feedback. Students bear major responsibility for the thesis’s outcomes.

Coach

The supervisor permits students to work alone coaching sessions are available upon request by students. The outcomes are mostly the responsibility of the students

adapted from [9]

the thesis is perhaps the most complex and challenging aspect of education in which faculty members are involved [19]. According to [26], the complicated interaction between students and supervisors stems from three different types of supervision: (1) traditional-academic, (2) technologicalscientific, and (3) psychological. Supervisors, according to [9], may be classified into four types, as illustrated in Table 1. As previously stated, a bachelor’s or master’s thesis is regarded as the students’ final preparation for professional careers. However, we have a few cases at Innopolis University who consider thesis completion a challenge while having completed all other required components of their study. Based on the conducted interviews with students and supervisors, we will identify the reasons behind this in the next sections. The following Sect. 3 elaborates on the study design, Sect. 4 states students’ impressions and feedback conveyed via survey, Sect. 5 presents the findings from the interviews with supervisors and students, Sect. 6 discusses the main outcomes and corresponding gamified solution ideas. Finally Sect. 7 draws main conclusions.

3 Study Design The study participants are two groups: supervisors and supervisees. We interviewed faculty members who supervise students in Computer Science (CS) laboratories, during thesis writing, on the one hand, and supervisees, in the same academic setting, on the other hand. The rationale for targeting both groups is to obtain a holistic picture of the thesis supervision process; in addition to investigating the correlation between both the supervisor’s and supervisee’s perspectives towards the thesis supervision process. From the supervisee’s group, we collected qualitative and quantitative data from two classes. First, we interviewed 3 CS graduate students. The semi-structured interviews were conducted online via Zoom videoconferencing. The objective is

406

R. Marouf et al.

to collect data from students who might not have any restrictions to mention the challenges and provide detailed examples on the thesis supervision experience, since these students have completed their degree. Therefore, the risk of bias can be at a minimal level. Meanwhile, the second class of supervisees targets mainly the students who are currently working on their theses or are still doing higher studies in the same academic setting. The survey has been conducted and there are 2 respondents. We tried to reach as many students as possible and therefore survey is the most feasible approach in this regard. The survey is confidential and anonymous and has been distributed to students via Telegram messages and through a Learning Management System. To get the perspective from supervisors’ side, we interviewed 5 CS faculty members who are actively supervising bachelors’ theses. The following sections list the questions that have been examined in the supervisor and supervisees’ interviews.

3.1 Interview Questions for the Supervisor The following questions were asked in the interview with the supervisors. The interviews with five supervisors were in person, except for one that was conducted via Zoom Videoconferencing. The interviews took 30–45 min. 1. What does thesis supervision involve in your lab as an academic activity? 2. How many hours do students receive during the thesis supervision and what determines the frequency of meetings? Standard/modifications to adjust to students’ needs? 3. What format does the thesis supervision take, questions and answers via emails messages, meetings, online/offline what format exactly? 4. How much aware are students of these patterns? Is there a manual? 5. What are the main challenges that a supervisor can face during the thesis supervision? Any examples? 6. What suggestions do you recommend to address thesis supervision challenges? 7. Are students motivated enough during the process? What can students do to easify the thesis supervision process? 8. How does thesis supervision impact the thesis completion and the success or failure of the work? 9. What factors can make the thesis supervision a success or a failure?

3.2 Interview Questions for Supervisees/Students The students who underwent face-to-face interviews were asked the following questions. 1. What was missing and the presence of which would have been more helpful to have during the thesis supervision process?

Thesis Supervision in Computer Science—Challenges and a Gamified …

407

2. What would you have done differently in this respect (as a supervisee)? 3. What do you think can make the thesis writing easier and more successful? In other words, how can the thesis supervision be improved from your perspective as a university graduate? Further a survey has been conducted to reach broader audience of students being supervised. The questions and corresponding students’ responses are stated in Sect. 4.

4 Student’s Feedback and Data Interview We interviewed three postgraduate students. Since they have already completed the degree, we believe their insights can provide a different perspective as they can reflect on the experience from a more neutral standpoint. The interview responses are examined in the Sect. 5. Survey We received 22 survey responses from CS graduates and undergraduates who have been through the experience of thesis writing and they have received thesis supervision. The students belong to diverse cultures and a variety of CS laboratories. Moreover, the participants were from both genders; however, since the work does not aim to highlight the dissimilarities in theses’ experience, the survey did not request marking the participant’s gender. Survey Responses In this section we provide a summary of responses received from supervisees. The number in the brackets indicate the number of students (out of total 22) who provided the specified response. 1. How do you describe the meetings for the thesis supervision this year? The thesis supervision was described as timely and helpful by (10), timely by (4), helpful by (8), untimely by (2). Whereas none find the thesis supervision unhelpful. 2. What feelings do you have during the thesis supervision? (multiple option selection allowed) The feelings were satisfied (14), stressed (4) neutral (3), and unsatisfied (1). 3. Give at least one reason to your answer above. We received positive responses from the students when they elaborated on the supervision meetings. Students stated that they feel satisfied, receive guidance through these meetings, get their questions answered, receive support that increases their understanding of the thesis topic, and they are able to progress. Some students also highlighted that their meetings are not regular and they meet the supervisor whenever there is a need. On the contrary, a student stated that they do not feel that meetings are necessary as they would like to do everything on their own. One stated that their supervisor cannot clearly state what results are expected from the thesis.

408

4.

5.

6. 7.

8.

9.

R. Marouf et al.

Another student informed that their supervisor provides guidance, despite of that sometimes the student feels stressed because of the complexity of the thesis topic. How did you decide who your supervisor is? (single choice question) I emailed them (10), I had a list of potential supervisors and I contacted them all (7), I asked for help from the education department (1), has other ways (4). How much time it took to know who your supervisor is? (single choice question) One week or less (9), two weeks (5), three weeks (1), one month (2), and more than a month after the semester starts (5). Did you select a topic of your own interest? Yes (19) and No (3). What is the most helpful activity during the thesis supervision to facilitate your thesis writing? The students mentioned several activities that they found very useful during the thesis work. They emphasized the importance of becoming aware of papers and workshops related to their research topic. In addition to the supervisor’s feedback on the thesis write-up, peer review is stated to be helpful. In terms of most helpful activities, several students have talked about the thesis write-up review and supervisors guidance about related articles. What challenges did you face during the thesis supervision process? Give examples please. Students highlighted several issues including the confusion about the further steps in research work, time management, writer’s block, meeting schedule conflicts, pressure of weekly meetings, and difference in opinion (between the supervisor and the student) on the research approach. On the contrary, one student stated that they did not feel anything challenging while being supervised. From your experience, what do you suggest to make the thesis supervision more successful? The students provided suggestions that were aligned with the issues stated in the previous question. The suggestions include a clear indication of the thesis progress steps, an extended duration for completion, student collaboration, and regular meetings.

5 Two Dimensional Challenges—Supervisor and Students’ Insights On the other side of the coin, we interviewed five CS faculty members who are actively supervising undergraduate and graduate theses (see Sect. 3.1 for interview questions). In addition, interviews were conducted with three postgraduate students (questions stated in Sect. 3.2) who have undergone the thesis supervision experience. The findings from interviews with supervisors were contrasted against the mixed data collected from a survey and interviews with students in CS. The findings were categorized, as demonstrated by the supervisor’s and supervisees stated in

Thesis Supervision in Computer Science—Challenges and a Gamified …

409

section “Interview Questions for Supervisees and Students”) who have undergone the thesis supervision experience. The findings from interviews with supervisors are contrasted with the data collected from the survey and interviews with students. We categorize the findings as demonstrated by the supervisor’s and supervisee’s, and their experiences into the following themes. Student Motivation and Rationale for Thesis Writing One of the missing elements during the thesis supervision was “to understand the meaning for the necessity for writing the thesis and get motivated”, as reported by a supervisor. Similarly, from the perspective of a student, “the meaning for producing fruitful work appears to be missing. Students are not self-motivated”. Interestingly, “motivation” and “the meaning” for thesis writing are linked, and their absence is argued by both the supervisor and supervisee to represent a challenge for the supervision towards thesis completion. Moreover, self-motivation is viewed as a leading factor for success in the thesis writing process, as a supervisor postulated that “finding more meaning will lead to finding the motivation”. Supervisor’s Support and Supervisee’s Autonomy The supervisor’s availability is one of the main factors characterizing thesis supervision. “The very first thing students expect from me is my availability,” a supervisor reported. And this explains the disappointment expressed in the supervisee’s questionnaire when students highlighted that one of the challenges was the absence of the supervisor’s management in guiding them throughout the thesis writing process. On the other hand, the supervisor’s availability is highlighted in the survey, as students suggested the supervisor’s availability as a determining factor for the success of the thesis supervision process. Learner autonomy is also an essential paradigm during the thesis supervision process. As a result, students whose level of proficiency in thesis completion is not identified in this study request supervisor autonomy support. A student described the inability to have the supervisor’s immediate support as “frustrating”. Suggestions for Efficient Thesis Supervision 1. From supervisors’ perspectives: The supervisors suggest starting the process of matching supervisors to students earlier (before the thesis semester starts). In addition, student profiles should indicate information about their research interest, not just their grades and students should be encouraged to take notes during the meeting. 2. From supervisees’ perspectives: The students emphasized on the need of having possibilities to select from a range of thesis topics. Further, students highlighted that they would like to present their research work more frequently infront of broader audience, be supervised by more than one supervisor from the research specialty area, and can benefit from elaborative familiarization with all stages of thesis completion. The students also feel the need to be provided with exemplary thesis templates and guidance to adopt to the write-up tools such as latex.

410

R. Marouf et al.

6 Opportunities for Gamified Thesis Supervision This study focuses on the major challenges faced by supervisors and supervisees in CS, during the thesis supervision process. Therefore, familiarization of corresponding academic settings with these challenges can provide academics with the appropriate approach to addressing them. Furthermore, such awareness of these challenges can help students who will eventually complete the work successfully, optimize time and efforts, especially when thesis projects are system building ones. On the other hand, the underlined challenges are the compass to propose solutions that can reduce the challenges during the thesis supervision for students and supervisors. “The best ideas are coming from students,” a supervisor reported. The supervisor emphasized the importance of discussing students’ research ideas and areas of interest. “When I offer my ideas, I offer my bias because I already have a solution”. Thus, the supervisor chooses to begin with the students’ areas of interest. In relation to the survey results, 17 students stated that they chose their topics of interest. Therefore, such a tendency to encourage supervisees to select their own thesis topic seems to be promoted in the academic setting. According to one postgraduate student, “Thesis writing is boring, and it will be interesting to see how students will communicate with the system”. The findings show that students have high expectations of what the supervisor should do. Furthermore, expectations on what supervisors should do is a prominent element highlighted by supervisees. Hence, the feasibility of such expectations indicate that supervisors’ workload is worth investigating in the extended version of our work. We believe that introducing gamification in the thesis supervision process has the potential of making certain issues light. As an instance, we propose a gameplay scenario when students are looking for thesis supervisors. The student’s research interests and supervisor’s expertise are each represented as a card, the more matches lead to more points and students prefer that supervisor. This would require a system (preferably digital) where supervisors and students both have the opportunity to represent their interest as digital cards. Further, a reward system can be initiated where students get points based upon how regular they are in meetings. This might reduce the tension between the supervisor and the student as the supervisor would not need to push the student for meetings and student would see their progress in this area. Rewards can also be associated with the progress in the thesis writeup. To make it more interesting and fair, students can also be provided the opportunity to send reward points to supervisors whenever they feel satisfied with supervisor’s guidance. For this gamified system we prefer cooperative gameplay setting. As in game Hanabi [27], all players cooperate and receive collective points, the students and supervisors both have to play well as their collective rewards will decide which kind of trophy (from bronze, silver, and gold) they win at the thesis submission time.

Thesis Supervision in Computer Science—Challenges and a Gamified …

411

7 Conclusions This paper identifies challenges of CS thesis supervision process. We conducted interviews with both supervisors and students to understand their experiences and perspectives on supervision process. The data obtained highlights the importance of regular meetings and supervisor’s guidance in the areas of referring to relevant research articles and on thesis write-up. We also realized that both groups (supervisor and student) are trying their best to achieve good outcomes and most of the issues highlighted require a little more transparent communication between the both groups. For this reason, we believe that introducing gamification can make things smooth and both groups will be able to clearly see the quality of their deliverables during the supervision process. Gamification is a light-hearted approach to reduce challenges associated with timelines, and communicating the impressions on each others work. The gamified system is also much significant as post-covid we continue to work in hybrid mode [28]. The less in-person interactions must be somehow compensated with other ways that facilitate the flow of transparent communication among the supervisors and students. This work is in progress and we continue to collect data and work on designing the mechanics for the suitable gamified system.

References 1. Kotorov, I., Krasylnykova, Y., Zhdanov, P., Mazzara, M., Aslam, H., Akhmetgaraeva, E., Naumcheva, M., Brown, J.A.: Institutional commitment and leadership as prerequisites for successful comprehensive internationalization. In: Frontiers in Software Engineering: First International Conference, ICFSE 2021, Innopolis, Russia, June 17–18, 2021, Revised Selected Papers 1, pp. 1–11. Springer (2021) 2. Mazzara, M., Zhdanov, P., Bahrami, M.R., Aslam, H., Kotorov, I., Imam, M., Salem, H., Brown, J.A., Pletnev, R.: Education after covid-19. In: Smart and Sustainable Technology for Resilient Cities and Communities, pp. 193–207. Springer (2022) 3. Yuloskov, A., Bahrami, M.R., Mazzara, M., Kotorov, I.: Smart cities in Russia: current situation and insights for future development. Future Internet 13(10), 252 (2021) 4. George, G., Lahani, K.R., Puranam, P.: Shcho zminylosia? vplyv pandemii covid na poriadok dennyi doslidzhen upravlinnia tekhnolohiiamy ta innovatsiiamy [what has changed? the impact of the covid pandemic on the technology and innovation management research agenda]. J. Manag. Res. 57(8), 1754–1758 (2020) 5. Kotorov, I., Krasylnykova, Y., Zhdanov, P., Mazzara, M.: Finding the right understanding: twenty-first century university, globalization and internationalization. In: Frontiers in Software Engineering Education: First International Workshop, FISEE 2019, Villebrumier, France, November 11–13, 2019, Invited Papers 1, pp. 341–353. Springer (2020) 6. Burgos, D.: Radical Solutions and Open Science: An Open Approach to Boost Higher Education. Springer Nature (2020) 7. Lessing, N., Lessing, A.C.: The supervision of research for dissertations and theses. Acta Commercii 4(1), 73–87 (2004) 8. Heffernan, T.: Bourdieu and Higher Education: Life in the Modern University. Springer Nature (2022) 9. Schefer-Wenzl, S., Miladinovic, I.: A structured process for supervising students’ final theses and projects in computer science. Int. J. Adv. Corporate Learn. 15(1) (2022)

412

R. Marouf et al.

10. Karunaratne, T.: Blended supervision for thesis projects in higher education: a case study. Electron. J. e-Learn. 16(2), 79–90 (2018) 11. Host, M., Feldt, R., Luders, F.: Support for different roles in software engineering master’s thesis projects. IEEE Trans. Educ. 53(2), 288–296 (2009) 12. Ghadirian, L., Sayarifard, A., Majdzadeh, R., Rajabi, F., Yunesian, M.: Challenges for better thesis supervision. Med. J. Islamic Republic of Iran 28, 32 (2014) 13. Kotorov, I., Krasylnykova, Y., Zhdanov, P., Mazzara, M.: Inter-nationalization strategy of Innopolis University. In: Frontiers in Software Engineering Education: First International Workshop, FISEE 2019, Villebrumier, France, November 11–13, 2019, Invited Papers 1, pp. 327–340. Springer (2020) 14. Ruchina, A.V., Kuimova, M.V., Polyushko, D.A., Sentsov, A.E., Jin, Z.X.: The role of research work in the training of master students studying at technical university. Procedia Soc. Behav. Sci. 215, 98–101 (2015); International Education and Cross-Cultural Communication, Problems and Solutions, IECC 2015, 9–11 June 2015, Tomsk, Russia 15. Adelakun-Adeyemo, O.: The computer science final year project: a time to mentor 16. Olsson, B., Berndtsson, M., Lundell, B., Hansson, J.: Running research-oriented final year projects for CS and is students. In: Proceedings of the 34th SIGCSE Technical Symposium on Computer Science Education, pp. 79–83 (2003) 17. Agu, N., Odimegwu, C.O.: Doctoral dissertation supervision: identification and evaluation of models. Educ. Res. Int. 2014 (2014) 18. Stacey, E., Wendy Fountain: Student and supervisor perspectives in a computer-mediated research relationship. In: 18th Annual Conference of the Australian Society for Computers in Learning in Tertiary Education (ASCILITE). Melbourne, Australia. Citeseer (2001) 19. Bazrafkan, L., Yousefy, A., Amini, M., Yamani, N.: The journey of thesis supervisors from novice to expert: a grounded theory study. BMC Med. Educ. 19(1), 1–12 (2019) 20. Blaschke, P., Demel, J., Kotorov, I.: Innovation performance of small, medium-sized, and large enterprises in Czechia and Finland. In: Liberec Economic Forum, pp. 21–29. Technical University of Liberec (2021) 21. Stappenbelt, B., Basu, A.: Student-supervisor-university expectation alignment in the undergraduate engineering thesis (2019) 22. Almeatani, M., Alotaibi, H., Alasmari, E., Meccawy, M., Alghamdi, B.: Thesis supervision mobile system for enhancing student-supervisor communication (2019) 23. Kloos, C.D., Alario-Hoyos, C., Morales, M., Rocael, H.R., OJerez, Ó., Pérez-Sanagustín, M., Kotorov, I., Alejandra Recinos Fernández, I., Magdiel Oliva-Cordova, L., Solarte, M., et al.: Prof-xxi: teaching and learning centers to support the 21st century professor. In: 2021 World Engineering Education Forum/Global Engineering Deans Council (WEEF/GEDC), pp. 447– 454. IEEE (2021) 24. Kotorov, I., Pérez-Sanagustín, M., Mansilla, F., Krasylnykova, Y., Hadaou, F.T., Broisin, J.: Supporting the monitoring of institutional competency in learning innovation: the prof-xxi tool. In: 2022 XVII Latin American Conference on Learning Technologies (LACLO), pp. 01–08. IEEE (2022) 25. Pérez-Sanagustín, M., Kotorov, I., Teixeira, A., Mansilla, F., Broisin, J, Alario-Hoyos, C., Jerez, Ó., do Carmo Teixeira Pinto, M., García, B., Kloos, C.D., et al.: A competency framework for teaching and learning innovation centers for the 21st century: anticipazting the post-covid-19 age. Electronics 11(3), 413 (2022) 26. Grant, B.M.: Fighting for space in supervision: fantasies, fairytales, fictions and fallacies. Int. J. Q. Stud. Educ. 18(3), 337–354 (2005) 27. Bauza, A.: Hanabi. R&R Games Inc. (2010) 28. Ibrahim, S.K.S., Kamaruddin, S.M., Yunus, M.M., Hashim, H.: The covid-19 pandemic and its impact on gamification within ESL virtual classrooms: a literature review. J. Acad. Res. Bus. Soc. Sci. 11(12), 268–277 (2021)

Humanizing Zoom: Lessons from Higher Education in Qatar R. Bianchi, B. Yyelland, A. Weber, K. Kittaneh, Sara Mohammed, Aia Zaina, Afreena Niaz, Huda Muazzam, Selma Fejzullaj, and Lolwa Al-Thani

Abstract Synchronous Conferencing Software (SCS) such as Zoom can be a mixed blessing presenting several advantages and disadvantages (Akyildiz in Int. J. Technol. Educ. Sci. 4:322–334, 2020 [Tümen Akyildiz in Int. J. Technol. Educ. Sci. 4:322– 334, 2020]; Kittaneh et al. in INTED2022 Proceedings, IATED, 2022 [Kittaneh, K., Weber, A., Bianchi, R., Laws, S., Yyelland, B.: Student online engagement in Qatar during covid-19: survey of expert opinion. In: INTED2022 Proceedings, pp. 8195– 8195. IATED (2022)]). Zoom provides several tools such as synchronous text chat, screen-sharing, emojis and reactions, and recordings that are not normally available in traditional face-to-face classrooms (Katz and Kedem-Yemini in J. Inf. Technol. Case Appl. Res. 23:173–212, 2021 [Katz and Kedem-Yemini in J. Inf. Technol. Case Appl. Res. 23:173–212, 2021]). In Qatar, Zoom helped higher education students continue their education during disruptive periods such as the Covid-19 pandemic and the FIFA 2022 World Cup. To investigate the opportunities and challenges of online learning in Qatar, a multidisciplinary group of pedagogical researchers at Qatar’s Education City (EC) trained students to conduct 22 focus groups among their peers, identifying the benefits and drawbacks of Zoom during and after the pandemic. Using Sketch Engine corpus analysis software, student focus group data was coded and synthesized to reflect student experiences of Zoom-mediated classrooms. Key findings from the study indicated that while several students enjoyed the freedom and flexibility of meeting virtually afforded by Zoom, many others experienced a harsh disconnect from their instructors and peers as a result of attending Zoom-only classrooms. Further, monotonous lecture-based instruction with little incentive for students to interact or participate left students feeling unsatisfied and demotivated. The researchers advocate a proactive stance on the part of instructors

R. Bianchi (B) · B. Yyelland · K. Kittaneh · S. Mohammed · A. Zaina · A. Niaz · H. Muazzam · S. Fejzullaj · L. Al-Thani VCUarts Qatar, PO Box 8095, Doha, Qatar e-mail: [email protected] A. Weber Weill Cornell Medicine Qatar, PO Box 24811, Doha, Qatar © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5_39

413

414

R. Bianchi et al.

and students to maximize the potential of Zoom by humanizing the classroom experience through creating opportunities for small group interactions, between-class collaborative projects, and individualized meetings between faculty and students.

1 Introduction Zoom, a widespread synchronous conferencing software (SCS) has several advantages and disadvantages for teaching [1, 2]. Zoom’s features such as synchronous text chat, screen sharing, emojis and reactions, and recordings offer learning opportunities not present in face-to-face classrooms [3]. In Qatar, as elsewhere in the world, Zoom allowed for education to continue during the COVID-19 pandemic when inperson teaching was either impossible or unfeasible. To study the experience of Zoom-mediated education, a multidisciplinary group of pedagogical researchers at Qatar’s Education City (EC) trained students and alumni to conduct 22 focus groups among their peers, identifying the benefits and drawbacks of Zoom during and after the pandemic. Using Sketch Engine corpus analysis software, student focus group data was coded and synthesized to reflect student experiences of Zoom-mediated classrooms. Sketch Engine analyses revealed that Zoom-mediated classrooms were often demotivating and that student behavior such as punctuality and participation were often negatively impacted. While some students reported advantages such as ease of access to classroom learning via Zoom, many others experienced alienation from their instructors and peers in these online settings due to a lack of accountability or incentive for student classroom participation. Repetitive lecture-based instruction further exacerbated students’ feelings of dissatisfaction and demotivation. As a result, the researchers advocate a proactive stance on the part of instructors and students to maximize the potential of Zoom by humanizing the classroom experience through creating opportunities for small group interactions, between-class collaborative projects, and individualized meetings between faculty and students. These concrete actions foster closer relationships, not only between faculty and students, but also among students themselves, leading to a richer, more engaging Zoom-mediated learning environment for all.

2 Literature Review 2.1 Background Technology has often been seen as a barrier to forming meaningful relationships with others especially among youth. Nevertheless, during the COVID-19 pandemic, several technologies became essential to the carrying out of everyday activities such

Humanizing Zoom: Lessons from Higher Education in Qatar

415

as attending classes. In Qatar, several software packages came to dominate the educational sphere during the pandemic, most notably, Zoom and WebEx as synchronous conferencing software (SCS) and Canvas as a multi-purpose learning management system (LMS). The literature below highlights some aspects of the complex relationships between learner attitudes and behavior, social relationships, and online learning.

2.2 Student Ability to Self-Regulate Several researchers have established a connection between student self-regulation and student success in online learning contexts [4–9]. Cho and Heron [5] identify self-regulated learners as those able to set and monitor their own learning goals. Lasfeto and Ulfa [7] further highlight the tendency of self-regulated learners to form relationships with their instructors and classmates. Taken together, effective goal-setting and classroom relationship building lead to a more satisfying learning experience for students [4, 10].

2.3 Student Satisfaction Several researchers have highlighted the importance of students experiencing satisfaction in the learning process [5, 9, 11–13]. While Zhu et al. [9] note that students overall tend to experience higher levels of satisfaction in face-to-face environments, this is not the whole picture. For instance, Al-Jaber and Al-Ghamdi [14] observe that students’ perceptions of educational infrastructure are another key component of their experience of satisfaction in learning. Almusharraf and Khahro [11] identify this infrastructure in clear course objectives, adequate lesson preparation, and effective tutorials and evaluation tools. Yet, these researchers point out that face-to-face learning is not an absolute prerequisite of student satisfaction, noting that online learning that incorporates both synchronous and asynchronous activities can also lead to student satisfaction in learning. They similarly note that student agency in their learning, however, is a key component of this satisfaction. El-Sayad et al. [6] draw a connection between these feelings of satisfaction and student perceptions of engagement.

2.4 Student–Teacher Bonding A body of research has emerged suggesting that, in successful online learning environments, teacher-student bonding is essential. Ally’s recent six-country research project identified qualities in teachers that contribute to successful online learning,

416

R. Bianchi et al.

which include computer skills, the ability to self-teach and social skills [15]. More recent research corroborates these findings, drawing a clear link between teacher engagement and student self-regulation [6, 7] and student satisfaction in online learning environments [6] and in the Qatari context [13].

2.5 Student Motivation Related to the above-mentioned themes, student motivation cannot be overemphasized. In this connection, students’ perceptions of having a prepared teacher positively affects their motivation to learn in online classrooms [10]. Similarly, student motivation correlates with their perceptions of the usefulness of a course and the fairness of the instructor [9]. Zhu et al. [7, 9] also observed a link between student self-regulation and intrinsic motivation. Nevertheless, Bawadi et al. [4] note that personal tragedies during the pandemic, expectedly, had a negative impact on student motivation [12]. In this regard, Soubra et al. [13] observe that teachers that help their students to better prepare for learning and promote more social interaction were able to increase their students’ motivation.

3 Research Methods 3.1 Research Design Ethnographic in nature, the data for this study consisted of student and teacher narratives captured using peer-moderated focus groups, a method noted for its efficiency in collecting significant amounts of information in a relatively short period of time [16]. The resulting textual data were then analyzed both quantitatively and qualitatively using the popular, commercially available corpus linguistic software, Sketch Engine [17]. The quantitative analysis consisted of the identification of statistically salient keywords from the data set. The qualitative analysis, in contrast, involved making research-informed decisions about which keywords to further analyze based on their connection to the themes identified in the literature review above relating to student motivation in online contexts.

3.2 Procedure A total of twenty-two (22) student focus groups were conducted; 21 in English and one in Arabic over two distinct time periods. First, thirteen (13) focus groups were convened in December 2022. Next, nine (9) additional focus groups were convened

Humanizing Zoom: Lessons from Higher Education in Qatar

417

from July through October 2022. The focus groups were conducted by research assistants and student mentees. All focus groups were conducted online using Zoom and duly recorded and transcribed. Each session lasted approximately 60 min. Participants for the focus groups were chosen through convenience sampling from higher education institutions across Qatar. Of the 95 original student participants, 78 selfreported female, 15 self-reported as male and 2 chose not to self-report gender. Almost all participants identified as belonging to one of eight higher education institutions. The focus groups ranged in size from three to eight participants.

3.3 Data Analysis A mix of corpus-driven and corpus-based approaches were [18, 19]. First, frequency lists of keywords were generated in order to identify lexical items that appeared to be particularly salient throughout the focus group narratives. In this way, the five most frequent open-class lexical items were identified. These five keywords were then categorized into groups that indexed student experiences, feelings, and behaviors. In this regard, three keywords emerged that reflected these areas. These keywords were then further examined using concordances in order to provide fine-grained detail about their use in context. The three keywords were: “demotivated,” “punctuality,” and “demotivating,” (see Table 1).

4 Results On the surface, several patterns seem to be apparent here. First, some of the words are lexically and semantically related such as “demotivated” and “demotivating.” Second, behavior-related lexis is also apparent in the form of “punctuality.” Third, context-related lexis such as “covid” and “semester.” But as cautioned earlier, such words mean little unless the wider textual context in which they appear is examined. Thus, a second stage involved running word sketches, i.e., collocational analyses of each of these words in order to better understand how and in what context they were Table 1 Top 5 keywords Rank

Item

Frequency (focus)

Frequency (reference)

Relative frequency (focus)

1

Demotivated

39

2634

194.57582

2

Punctuality

39

19,277

194.57582

3

Covid

87

189,999

434.05377

4

Semester

278

777,555

1386.97644

5

Demotivating

15

2069

74.83685

418

R. Bianchi et al.

Fig. 1 Word sketch of “demotivated”

used. Figure 1 displays the word sketch of “demotivated”, which occurred a total of 39 times: Looking at modifiers of “demotivated,” “very demotivated” emerged in the following student observation: I felt like everybody was very demotivated, and at some point, I was as well, and I felt very fatigued that a lot of my day was not spent in uni, but rather spent at home. So, I don’t feel like I was doing any actual work. Clearly, this participant felt that being at home was not the same as being at university doing work. Another observation revealed both motivating and demotivating factors: I found what motivated me honestly was the way the classes taught, how easygoing it was and the material itself that motivated me to stay in class. The factor that really demotivated me was the lack of sleep, even though like not only that, it was the lack of communication, that sometimes I would be really demotivated to enter class, even though I had, like, flexible professors and really kind ones, I was sometimes like I would get bothered. Like, “why are they giving us this many assignments?”. Here two reasons for demotivation are given, a lack of sleep and an overabundance of assignments. However, another respondent gave the cause of demotivation as a lack of interaction and not knowing others as someone new to the university: What demotivated me would be basically the lack of interaction because you are new to uni. It is your first time at uni, you don’t know anyone…[inaudible] that demotivated a little bit. Another respondent, contrasting the person who had just spoken, mentioned that unsatisfying teaching was demotivating and that some online classes were “just OK” and therefore it was harder to concentrate in these classes and want to attend them. However, when these classes moved to in-person teaching, the same person was eager to meet their classmates in person: That was just the motivating factor. I was demotivated because again, like the teaching was just ‘meh’ during the online class, they were just OK. It was just OK. It was definitely harder for me to concentrate during those classes, so I was demotivated to go for them. But I was still intrigued to know who was actually in my classes and who I’d be seeing in person that we went when we stopped being online. In total, “demotivating” occurred 15 times. A review of the concordance lines highlighted several situations or elements of online learning that students described as “demotivating.” Some of the issues discussed were: the monotony of the teaching approach of certain online classes and the routine of having to be online every day; the lack of in-class on-task accountability due to student cameras being off; the

Humanizing Zoom: Lessons from Higher Education in Qatar

419

Fig. 2 Word sketch of “Punctuality”

heavy course workload; teacher favoritism toward specific students; and the frequent interruptions of family members when attending class online and at home. At this point, it is worth mentioning that another semantically-related word, “demotivation”, while not in the original list of five keywords because it occurred slightly less frequently, was, nevertheless, a common item, occurring 14 times i.e., just once less than “demotivating.” In connection with “demotivation”, the following points were observed in the student narratives: a lack of connection with instructors and classmates due to not having to have cameras on; certain students dominating class discussions and “asking all the questions”; interactive lessons as remedying demotivation; and too much freedom and/or a lack of accountability for online in-class behavior. Within the five keywords, another frequent item was “punctuality,” which occurred 39 times in the corpus. And while most of these were instances of the focus group moderator’s questions, some occurred as part of the participants’ answers. In this regard, in Fig. 2, several patterns can be traced. First, punctuality has been personalized as it is often modified by “my,” “your,” or “our.” Next, punctuality is seen as something that has improved or increased. Third, punctuality clusters with other classroom-related issues such as “attendance”, “participation.” Investigating these further, one participant comment was very interesting, indicating that while her punctuality improved due to the ease of attending online, there were several days when she was not engaged at all in the learning process due to a lack of accountability in the online classroom: I think I mentioned at the beginning that I did feel like my punctuality improved and my attendance improved because I always had trouble getting up and getting to class on time and I had to get ready and just be there physically. But there were days, for example, where you would be lazy or for example, on days that I overslept and I didn’t get to do my whole routine. And I was very sleepy and I just woke up right in time for class. I wouldn’t even open on my laptop. I would just like to have my phone beside me. So, I would just open Zoom on my phone and I would have my camera and microphone off and just be in bed taking class and like half asleep. And, you know, I mean, because there was no kind of accountability. Meanwhile, another respondent reported the following when asked about how their punctuality had changed when returning to in-person classes: Um, I think my punctuality has increased because I’m actually, I actually have a reason to go. Zoom classes before they didn’t really take attendance so that was and

420

R. Bianchi et al.

it was easy to say, “Oh, I had connection issues” if you were not in class. So now you can’t really see my connection affecting my attendance. Both of these comments appear to speak to a common theme of a lack of accountability in online classes impacting the quality of student participation.

5 Discussion The keyword analyses above highlight the extent to which students experienced demotivation in online classrooms. This demotivation was due to several factors ranging from boring classrooms to heavy workloads to a lack of accountability for student participation. This last issue was seen in large part as a consequence of institutional policies that allowed students to turn off their cameras and microphones during Zoom-mediated lessons. Punctuality was also discussed frequently in classrooms, and while ease of attendance was clearly noted as students merely had to log on to attend a class on time, clearly, such punctuality was not synonymous with meaningful online participation and engagement when such participation was not incentivized in any meaningful way.

5.1 Limitations Several limitations to the research should be noted. First, since focus group participants were not drawn from a random sampling of all of Qatar, the findings cannot be considered generalizable beyond this study. Additionally, since the population reflected a strong female bias, future research with a more equally distributed gender ratio would likely reveal different results. Third, while the study utilizes sophisticated discourse-based software to analyze the data, the thick data associated with traditional ethnography are not included here, although this would be valuable for future research. Lastly, no attempt was made in the current study to investigate and analyze class, race, religion or cultural differences amongst focus group participants. Each of these social factors could be examined in future research.

6 Conclusions The findings in this study contribute to the extant body of knowledge in four key areas. First, findings from the study indicated that several students enjoyed the freedom and flexibility of meeting virtually afforded by Zoom. However, the second finding is that many other students experienced a harsh disconnect from their instructors and peers as a result of attending Zoom-only classrooms in which they were allowed to turn off

Humanizing Zoom: Lessons from Higher Education in Qatar

421

their cameras and mute their microphones. Third, pedagogical processes and strategies became paramount within the online environment. Monotonous lecture-based instruction with little incentive for students to interact or participate left students feeling unsatisfied and demotivated, directly impacting feelings of engagement. Finally, informed by these data, the authors advocate instructors and students adopt a proactive stance to maximize the potential of Zoom through humanizing the classroom experience. In order to achieve this, in addition to mandating camera and mic use in order to encourage both participation and accountability, other methods of humanization include creating opportunities for small group interactions both during classes and between-class collaborative projects as well as setting up individualized meetings between faculty and students outside of formal class meetings. As evidenced from the student focus group narratives, such measures will undoubtedly lead to greater rapport among both students and instructors. Additionally, encouraging students to connect with each other through other social media platforms such as Instagram and WhatsApp can further promote student-to-student engagement and rapport. In summary, the researchers note that these concrete actions seem to foster closer relationships, not only between faculty and students, but also among students themselves, leading to a richer, more engaging Zoom-mediated learning environment for all. Acknowledgements This research was made possible in part by funding from the Undergraduate Research Experience Program (UREP) provided by the Qatar National Research Fund (QNRF), a Faculty Research Grant provided by Virginia Commonwealth University Qatar (VCUarts Qatar), and a Multiversity Grant from Qatar Foundation (QF). The authors wish to acknowledge the efforts of Sara Mohammed, Aia Zaina, Afreena Niaz, Huda Muazzam, Lolwa Al-Thani, and Selma Fejzullaj in researching and preparing this article for publication. Conflict of Interest The authors declare they have no conflicts of interest. Ethical Approval . Informed Consen The research protocol meets the requirements of the National Statement on Ethical Conduct in Human Research. Ethical approval for this research was granted by the Institutional Research Boards (IRBs) of Hamad bin Khalifa University and Weill Cornell University under IRB 2020-11-037 and 20000-40 respectively. All participants received and signed an approved Informed Consent form prior to participation in the study.

References 1. Tümen Akyildiz, S.: College students’ views on the pandemic distance education: a focus group discussion. Int. J. Technol. Educ. Sci. 4(4), 322–334 (2020) 2. Kittaneh, K., Weber, A., Bianchi, R., Laws, S., Yyelland, B.: Student online engagement in Qatar during covid-19: survey of expert opinion. In: INTED2022 Proceedings, pp. 8195–8195. IATED (2022)

422

R. Bianchi et al.

3. Katz, A., Kedem-Yemini, S.: From classrooms to zoom rooms: preserving effective communication in distance education. J. Inf. Technol. Case Appl. Res. 23(3), 173–212 (2021) 4. Bawadi, H., Rahim, H.A., Moawad, J., Shami, R., Xiangyun, D., El-Awaisi, A., Al-Moslih, A.M.I., Diab, M., Al-Jayyousi, G.F.: Health science students’ and instructors’ perceptions of the emergency switch to virtual internship amid the covid-19 pandemic: a case from Qatar. Front. Med. 9, 939416 (2022) 5. Cho, M.H., Heron, M.L.: Self-regulated learning: the role of motivation, emotion, and use of learning strategies in students’ learning experiences in a self-paced online mathematics course. Distance Educ. 36(1), 80–99 (2015) 6. El-Sayad, G., Saad, N.H.M., Thurasamy, R.: How higher education students in Egypt perceived online learning engagement and satisfaction during the COVID-19 pandemic. J. Comput. Educ. 8(4), 527–550 (2021) 7. Lasfeto, D.B., Ulfa, S.: The relationship between self-directed learning and students’ social interaction in the online learning environment. J. E-Learn. Knowl. Soc. 16(2), 34–41 (2020) 8. Park, S., Kim, S.: Is sustainable online learning possible with gamification? - The effect of gamified online learning on student learning. Sustainability 13, 4267 (2021) 9. Zhu, Y., Zhang, J.H., Yates, G.: University students’ online learning attitudes and continuous intention to undertake online courses: a self-regulated learning perspective. Educ. Technol. Res. Dev. 68, 1485–1519 (2020) 10. El-Masri, M., Tarhini, A.: Factors affecting the adoption of e-learning systems in Qatar and USA: extending the unified theory of acceptance and use of technology 2 (UTAT2). Educ. Technol. Res. Dev. 65, 743–763 (2017) 11. Almusharraf, N.M., Khahro, S.H.: Students; satisfaction with online learning experiences during the COVID-19 pandemic. Int. J. Emerg. Technol. 15(21), 2426–267 (2020) 12. Karim, Z., Javed, A., Azeem, M.W.: The effects of Covid-19 on medical education. Pakistan J. Med. Sci. 38(1), 320–322 (2022) 13. Soubra, L., Al-Ghourti, M.A., Abu-Dieyeh, M., Crovella, S., Abou-Saleh, H.: Impacts on student learning and skills implementation challenges of two student-centered learning methods applied in online education. Sustainability 14, 9625 (2022) 14. Al-Jaber, M.A., Al-Ghamdi, S.G.: Effect of virtual learning on delivering the education as part of the sustainable development goals in Qatar. Sci. Direct Energy Rep. 6, 371–375 (2020) 15. Ally, M.: Competency profile of the digital and online teacher in future education. Int. Rev. Res. Open Distrib. Learn. 20(2), 302–318 (2019) 16. Wilson, V.: Research methods: focus groups. Evid. Based Libr. Inf. Pract. 11(1), 44–46 (2016) 17. Kilgarriff, A., Baisa, V., Bušta, J., Jakubíˇcek, M., Kováˇr, V., Michelfeit, J., Rychlý, P., Suchomel, V.: The sketch engine: ten years on. Lexicography 1, 7–36 (2014) 18. Baker, P., Gabrielatos, C., McEnery, T.: Sketching Muslims: a corpus driven analysis of representations around the word ‘Muslim’ in the British Press 1998–2009. Appl. Linguis. 34(3), 255–278 (2012) 19. Tognini-Bonelli, E.: Corpus Linguistics at Work. John Benjamins (2001)

Author Index

A Abe, Kenta, 41 Abu, Patricia Angela R., 221 Ahmed, El Darawany, 233 Alanis, Arnulfo, 241, 251 Alisoltani, Negin, 119 Alqasir, Hiba, 15 Al-Thani, Lolwa, 413 Angel Casillas-Araiza, Miguel, 251 Anureev, I. S., 373 Ascención Guerrero-Viramontes, J., 241 Aslam, Hamna, 403

B Badeig, Fabien, 15 Bahrami, Bahareh, 393 Bahrami, Mohammad Reza, 277, 393 Balbo, Flavien, 15 Baltazar, Rosario, 241, 251 Batton-Hubert, Mireille, 15 Behboodi, Farima, 393 Bek, Burak, 99 Bianchi, R., 413 Bobrov, Evgeny, 383

C Campagna, Swen, 3 ˇ Cavrak, Igor, 3 Ceobanu, Marius Ciprian, 313 Cherfi, Anis, 55 Ciancarini, Paolo, 289

D Daoud, Alaa, 15 Deffner, Demian, 161 Demel, Jaroslav, 383 Distefano, Salvatore, 355 Driss, Olfa Belkahla, 129

E Ehara, Takashi, 183

F Fejzullaj, Selma, 413 Francisco-Mosiño, J., 251

G Gaussier, Natalie, 87 Gazem, Miriam, 301 Ghedira, Khaled, 129 Gilb, Tom, 339 Glass, Ayse, 99 González-Mota, R., 241 Gorecki, Simon, 87

H Härting, Ralf-Christian, 161, 171, 301 Hattab, Siham Siham, 333 Halaška, Michal, 109 Hayashi, Hisashi, 41 Hrytsuk, Yury, 323

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 G. Jezic et al. (eds.), Agents and Multi-Agent Systems: Technologies and Applications 2023, Smart Innovation, Systems and Technologies 354, https://doi.org/10.1007/978-981-99-3068-5

423

424 I Imbugwa, Gerald B., 339 Inagaki, Hitomi, 151 Ishino, Yoko, 207 Islam, Robiul, 277

J Jezic, Gordan, 65 Johard, Leonard, 277 Julia, Wirth, 233 Juganaru-Mathieu, Mihaela, 15

K Kadriu, Festina, 171 Karg, Christoph, 161 Kittaneh, K., 413 Kladko, Sergei, 265 Kleemann, Julia, 171 Kondratyev, D. A., 373 Kotorov, Iouri, 383, 403 Kovaˇcevi´c, Marko, 27 Kovaˇci´c, Zdenko, 27 Kralj, Ivan, 65 Kurahashi, Setsuya, 141, 151 Kusek, Mario, 65 Krasylnykova, Yuliya, 383, 403 Kudasov, N., 373

M Magboo, Vincent Peter C., 221 Marina, Kholod, 233 Marina, Shilina, 233 Marouf, Rabab, 403 Marzouki, Bilel, 129 Matsui, Minoru, 41 Mazzara, Manuel, 383 Missiroli, Marcello, 289 Mohammed, Sara, 413 Muazzam, Huda, 413 Muller, Guillaume, 15

N Nakamura, Hideto, 207 Naumcheva, Maria, 347 Niaz, Afreena, 413 Noennig, Jörg Rainer, 99 Nouira, Kaouther, 55

Author Index P Pezer, Marko, 403 Pourrafie, Samae, 393

R Reichstein, Christopher, 171, 301 Rybakov, Vladimir V., 79

S Salem, Hamza, 333 Schneider, Anna, 301 Schulz, Greg-Norman, 161 Scutelnicu, Liviu-Andrei, 313 Shilov, N. V., 373 Sorg, Natalie, 171 Soto-Bernal, Juan José, 241 Šperka, Roman, 109 Steffens, Ulrike, 99 Stranjak, Armin, 3

T Takahashi, Hiroshi, 183, 195 Thapaliya, Ananga, 323 Tokuç, Kübra, 99 Traore, Mamadou Kaba, 87

V Vera, Ángeles Arellano, 241 Vilchis, Gerardo, 251

W Weber, A., 413

Y Yashima, Kohei, 141 Yamamoto, Daishiro, 195 Yyelland, B., 413

Z Zaina, Aia, 413 Zaman, Mustafeed, 277 Zargayouna, Mahdi, 119 Zerguini, Seghir, 87 Zlatanov, Nikola, 367