Trends and Applications in Information Systems and Technologies: Volume 4 (Advances in Intelligent Systems and Computing, 1368) 3030726533, 9783030726539

Citation preview

Advances in Intelligent Systems and Computing 1368

Álvaro Rocha · Hojjat Adeli · Gintautas Dzemyda · Fernando Moreira · Ana Maria Ramalho Correia   Editors

Trends and Applications in Information Systems and Technologies Volume 4

Advances in Intelligent Systems and Computing Volume 1368

Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Advisory Editors Nikhil R. Pal, Indian Statistical Institute, Kolkata, India Rafael Bello Perez, Faculty of Mathematics, Physics and Computing, Universidad Central de Las Villas, Santa Clara, Cuba Emilio S. Corchado, University of Salamanca, Salamanca, Spain Hani Hagras, School of Computer Science and Electronic Engineering, University of Essex, Colchester, UK László T. Kóczy, Department of Automation, Széchenyi István University, Gyor, Hungary Vladik Kreinovich, Department of Computer Science, University of Texas at El Paso, El Paso, TX, USA Chin-Teng Lin, Department of Electrical Engineering, National Chiao Tung University, Hsinchu, Taiwan Jie Lu, Faculty of Engineering and Information Technology, University of Technology Sydney, Sydney, NSW, Australia Patricia Melin, Graduate Program of Computer Science, Tijuana Institute of Technology, Tijuana, Mexico Nadia Nedjah, Department of Electronics Engineering, University of Rio de Janeiro, Rio de Janeiro, Brazil Ngoc Thanh Nguyen , Faculty of Computer Science and Management, Wrocław University of Technology, Wrocław, Poland Jun Wang, Department of Mechanical and Automation Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong

The series “Advances in Intelligent Systems and Computing” contains publications on theory, applications, and design methods of Intelligent Systems and Intelligent Computing. Virtually all disciplines such as engineering, natural sciences, computer and information science, ICT, economics, business, e-commerce, environment, healthcare, life science are covered. The list of topics spans all the areas of modern intelligent systems and computing such as: computational intelligence, soft computing including neural networks, fuzzy systems, evolutionary computing and the fusion of these paradigms, social intelligence, ambient intelligence, computational neuroscience, artificial life, virtual worlds and society, cognitive science and systems, Perception and Vision, DNA and immune based systems, self-organizing and adaptive systems, e-Learning and teaching, human-centered and human-centric computing, recommender systems, intelligent control, robotics and mechatronics including human-machine teaming, knowledge-based paradigms, learning paradigms, machine ethics, intelligent data analysis, knowledge management, intelligent agents, intelligent decision making and support, intelligent network security, trust management, interactive entertainment, Web intelligence and multimedia. The publications within “Advances in Intelligent Systems and Computing” are primarily proceedings of important conferences, symposia and congresses. They cover significant recent developments in the field, both of a foundational and applicable character. An important characteristic feature of the series is the short publication time and world-wide distribution. This permits a rapid and broad dissemination of research results. Indexed by DBLP, EI Compendex, INSPEC, WTI Frankfurt eG, zbMATH, Japanese Science and Technology Agency (JST), SCImago. All books published in the series are submitted for consideration in Web of Science.

More information about this series at http://www.springer.com/series/11156

Álvaro Rocha Hojjat Adeli Gintautas Dzemyda Fernando Moreira Ana Maria Ramalho Correia •

•

•

•

Editors

Trends and Applications in Information Systems and Technologies Volume 4

123

Editors Álvaro Rocha ISEG University of Lisbon Lisbon, Portugal Gintautas Dzemyda Institute of Data Science and Digital Technologies Vilnius University Vilnius, Lithuania

Hojjat Adeli College of Engineering The Ohio State University Columbus, OH, USA Fernando Moreira DCT Universidade Portucalense Porto, Portugal

Ana Maria Ramalho Correia Department of Information Sciences University of Sheffield Lisbon, Portugal

ISSN 2194-5357 ISSN 2194-5365 (electronic) Advances in Intelligent Systems and Computing ISBN 978-3-030-72653-9 ISBN 978-3-030-72654-6 (eBook) https://doi.org/10.1007/978-3-030-72654-6 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2021, corrected publication 2022 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 Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

This book contains a selection of papers accepted for presentation and discussion at the 2021 World Conference on Information Systems and Technologies (WorldCIST’21). This conference had the scientific support of the University of Azores, Information and Technology Management Association (ITMA), IEEE Systems, Man, and Cybernetics Society (IEEE SMC), Iberian Association for Information Systems and Technologies (AISTI), and Global Institute for IT Management (GIIM). It took place online at Hangra do Heroismo city, Terceira Island, Azores, Portugal, March 30–31 to April 1–2, 2021. The World Conference on Information Systems and Technologies (WorldCIST) is a global forum for researchers and practitioners to present and discuss recent results and innovations, current trends, professional experiences, and challenges of modern information systems and technologies research, technological development, and applications. One of its main aims is to strengthen the drive toward a holistic symbiosis between academy, society, and industry. WorldCIST’21 built on the successes of WorldCIST’13 held at Olhão, Algarve, Portugal; WorldCIST’14 held at Funchal, Madeira, Portugal; WorldCIST’15 held at São Miguel, Azores, Portugal; WorldCIST’16 held at Recife, Pernambuco, Brazil; WorldCIST’17 held at Porto Santo, Madeira, Portugal; WorldCIST’18 held at Naples, Italy; WorldCIST’19 held at La Toja, Spain; and WorldCIST’20, which took place online at Budva, Montenegro. The Program Committee of WorldCIST’21 was composed of a multidisciplinary group of 309 experts and those who are intimately concerned with information systems and technologies. They have had the responsibility for evaluating, in a ‘blind review’ process, the papers received for each of the main themes proposed for the conference: A) information and knowledge management; B) organizational models and information systems; C) software and systems modeling; D) software systems, architectures, applications and tools; E) multimedia systems and applications; F) computer networks, mobility and pervasive systems; G) intelligent and decision support systems; H) big data analytics and Applications; I) human–computer interaction; J) ethics, computers and security; K) health informatics;

v

vi

Preface

L) information technologies in education; M) information technologies in radiocommunications; N) technologies for biomedical applications. The conference also included workshop sessions taking place in parallel with the conference ones. Workshop sessions covered themes such as healthcare information systems interoperability, security and efficiency; user expression and sentiment analysis; gamification application and technologies; code quality and security; amalgamating artificial intelligence and business innovation; innovation and digital transformation for rural development; automatic detection of fake news in social media; open learning and inclusive education through information and communication technology; digital technologies and teaching innovations in COVID-19 times; devops and software engineering; pervasive information systems; advancing eHealth through software engineering fundamentals; blockchain and distributed ledger technology (DLT) in business; innovation and intelligence in educational technology, evolutionary computing for health care; ICT for auditing and accounting; and leveraging customer behavior using advanced data analytics and machine learning techniques. WorldCIST’21 received about 400 contributions from 51 countries around the world. The papers accepted for oral presentation and discussion at the conference are published by Springer (this book) in four volumes and will be submitted for indexing by WoS, EI-Compendex, Scopus, DBLP, and/or Google Scholar, among others. Extended versions of selected best papers will be published in special or regular issues of relevant journals, mainly JCR/SCI/SSCI and Scopus/EI-Compendex indexed journals. We acknowledge all of those that contributed to the staging of WorldCIST’21 (authors, committees, workshop organizers, and sponsors). We deeply appreciate their involvement and support that was crucial for the success of WorldCIST’21. March 2021

Álvaro Rocha Hojjat Adeli Gintautas Dzemyda Fernando Moreira

Contents

Software and Systems Modeling Analysis of German National Electricity Grid at Risk of Random Damage - Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Piotr Hadaj and Dominik Strzałka

3

Advanced Smart Car Park Management System for High Car Park Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ammar Sami Aldallal, Aysha Al Zallaqi, and Amina Mohammed

11

Analyzing and Modeling Critical Risks in Software Development Projects: A Study Based on RFMEA and Systems Dynamics . . . . . . . . . Lucilene da Silva Leite, Alderedo Rubem de Farias Neto, Fernanda Lopes de Lima, and Ricardo Matos Chaim The Communication Between Client-Developer in the Process of Requirements Elicitation for a Software Project . . . . . . . . . . . . . . . . Sebastián Alvarez, Kevin Duy, Mireya Zapata, Jorge Galarza, Danilo Martinez, and Carlos Puco

22

36

Unreadable Code in Novice Developers . . . . . . . . . . . . . . . . . . . . . . . . . Daniel Avila, Edison Báez, Mireya Zapata, David López, Diego Zurita, and Danilo Martínez

46

Security Smart Farming System – AgriPro . . . . . . . . . . . . . . . . . . . . . . Janani Tharmaseelan, S. V. Aaran, D. Ravipprasath, and P. Jayapragash

52

Smart Contract Code Generation from Platform Specific Model for Hyperledger Go . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mantas Jurgelaitis, Vaidotas Drungilas, Lina Čeponienė, Evaldas Vaičiukynas, Rita Butkienė, and Jonas Čeponis A Scoping Definition Experiment in Software Process Lines . . . . . . . . . Pablo H. Ruiz, Vanessa Agredo-Delgado, Alicia Mon, Cesar A. Collazos, Fernando Moreira, and Julio A. Hurtado

63

74

vii

viii

Contents

Best Practices for OutSystems Development and Its Influence on Test Automation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Joana Salgueiro, Fernando Ribeiro, and José Metrôlho

85

Solving the Multi-objective 2-Dimensional Vector Packing Problem Using
-constraint Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nadia Dahmani, Saoussen Krichen, El-Ghazali Talbi, and Sanaa Kaddoura

96

A Systematic Review of Risk Management in System Dynamics Project and Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 Eliane Cunha Marques, Marcos Vinicius L. Castro, Hilario Luiz Babireski Junior, and Ricardo Matos Chaim A Brokering-Based Interaction Protocol for Dynamic Resource Allocation in Steel Production Processes . . . . . . . . . . . . . . . . . . . . . . . . 119 Vincenzo Iannino, Claudio Mocci, and Valentina Colla A Tool to Support the Investigation and Visualization of Cyber and/or Physical Incidents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 Inês Macedo, Sinan Wanous, Nuno Oliveira, Orlando Sousa, and Isabel Praça Use of Virtual Reality as Learning Object of Electrical Substations . . . 141 Alexandre Carvalho Silva, Arnaldo J. P. Rosentino Jr., Alexandre Cardoso, and Edgard Afonso Lamounier Jr. Stewart Platform Modified into a Bio-inspirational Device of the Human Shoulder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 Marius Leonard Olar, Arun Fabian Panaite, Monica Leba, and Remus Sibișanu Multimedia Systems and Applications The Design of Social Coupons Used as a Digital Promotional Tactic . . . 163 Joana Sampaio Correia and Dora Simões A Tracking Topology to Support a Trajectory Monitoring System Directed at Helping Visually Impaired on Paralympic Athletics . . . . . . 173 Luiz Alberto Queiroz Cordovil Júnior, Renan Lima Baima, Gabriel Tadayoshi Rodrigues Oka, Ana Carolina Oliveira Lima, Moisés Pereira Bastos, Ana Isabel Martins, and Nelson Pacheco Rocha Physical Movement Helps Learning: Teaching Using Tracking Objects with Depth Camera . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 André Sales Mendes, Luis Augusto Silva, Héctor Sánchez San Blas, Daniel H. de La Iglesia, Francisco García Encinas, Valderi Reis Quietinho Leithardt, and Gabriel Villarubia González

Contents

ix

Authoring Tool for the Creation of Learning Objects in Occupational Education to Support People with Cognitive Disabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 Paola Pachacama, Luis Cuascota, and Graciela Guerrero Home Assistance Technologies for Elderly People: A Brief Evaluation and Software Architectural Proposal . . . . . . . . . . . . . . . . . . 205 Juan P. D‘Amato, Mauricio González-Palacio, Alejandro Perez, Leonardo Dominguez, Aldo Rubiales, and Franco Stramana Conversational Agent for Industrial Processes Through Virtual Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218 Francisco I. Teneda, Johnny I. Villacís, Edison G. Espinosa, and Víctor H. Andaluz Technologies for Biomedical Applications Tissue-Mimicking Phantoms: Dielectric Characterization and Design of a Multi-layer Substrate for Microwave Blood Glucose Monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Sandra Costanzo, Vincenzo Cioffi, and Giuseppe Lopez Innovative Equipment to Monitor and Control Salt Usage When Cooking: Acceptance of This New Technology to Healthy Eating Promotion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240 Carla Gonçalves, Tânia Silva-Santos, Patrícia Padrão, Pedro Moreira, Sílvia Esteves, Luís Oliveira, and Olívia Pinho Dielectric Characterization of Breast Phantoms for Phaseless Microwave Imaging Focused on Tumor Identification . . . . . . . . . . . . . . 248 Sandra Costanzo and Giuseppe Lopez IoT Non-contact Body Temperature Measurement System Implementing Access Control for COVID-19 . . . . . . . . . . . . . . . . . . . . . 255 Sandra Costanzo and Alexandra Macarena Flores Miscellaneous from Workshops Gamification in MOOC Context. Documentary Analysis . . . . . . . . . . . . 267 Angel Torres-Toukoumidis, Sonia Esther González-Moreno, and Jesús Manuel Palma-Ruiz RETRACTED CHAPTER: Semantic Analysis on Twitter of Two Religious’ Leaders: @Pontifex and @DalaiLama . . . . . . . . . . . . 278 Angel Torres-Toukoumidis, Isidro Marín-Gutiérrez, and Javier Herrán Gómez

x

Contents

Step Towards Predicting Patient Length of Stay in Intensive Care Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287 Cristiana Neto, Gabriel Pontes, Alexandru Domente, Francisco Reinolds, José Costa, Diana Ferreira, and José Machado Numerical Validation of Half Precision Simulations . . . . . . . . . . . . . . . . 298 Fabienne Jézéquel, Sara sadat Hoseininasab, and Thibault Hilaire A Step Towards the Use of Chatbots to Support the Enterprise Decision-Making Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308 Diogo Ferreira, Filipe Portela, and Manuel Filipe Santos COVID-19’s Impact on the Telecommunications Companies . . . . . . . . . 318 Latifah Almuqren and Alexandra I. Cristea Link Prediction Using Double Degree Equation with Mutual and Popular Nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328 Haji Gul, Adnan Amin, Furqan Nasir, Sher Jeel Ahmad, and Muhammad Wasim Trends in the Use of 3D Printing with Medical Imaging . . . . . . . . . . . . 338 Tiago Jesus and Victor Alves Paper-Code-Az: Steps for Assembling a Paper-Code for an Educational Robotic System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348 Matthias Funk, José Cascalho, Michael Funk, Paulo Novo, Francisco Pedro, and Armando Mendes A Practical Solution to Synchronise Structured and Non-structured Repositories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356 Vanessa Ferreira, Filipe Portela, and Manuel Filipe Santos Lesson Learnt by Using DevOps and Scrum for Development a Traceability Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366 Dimitrios Salmas, Giannis Botilias, Jeries Besharat, and Chrysostomos Stylios Social Actions and Corporate Social Responsibility in the Public Service Media of Spain, France and Portugal . . . . . . . . . . . . . . . . . . . . 374 Clide Rodríguez-Vázquez, María-Magdalena Rodríguez-Fernández, Tania Fernández-Lombao, and Matías Membiela-Pollán Virtual Reality to Improve Postural Control in Patients with Schizophrenia: Study Protocol for a Single-Blind Parallel Group Randomised Controlled Trial . . . . . . . . . . . . . . . . . . . . . . . . . . . 384 Mara Cunha, Raquel Simões de Almeida, Ângela Fernandes, and António Marques

Contents

xi

Decision Support Platform for Intelligent and Sustainable Farming . . . 401 Denisa Pastea, Daniela-Marina Draghici, George Suciu, Mihaela Balanescu, George-Valentin Iordache, Andreea-Geanina Vintila, Alexandru Vulpe, Marius Vochin, Ana-Maria Dragulinescu, and Catalina Dana Popa Digitally Monitoring Thermalism Health and Wellness Effects A Conceptual Model Proposal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411 José Martins, Fernando Moreira, Manuel Au-Yong-Oliveira, Ramiro Gonçalves, and Frederico Branco E-commerce During the COVID-19 Pandemic: An Exploratory Study Involving Portuguese Consumers . . . . . . . . . . . . . . . . . . . . . . . . . 422 Bruna Plácido, Sofia Proença, Diogo Moreira, Lorena Boução, Frederico Branco, and Manuel Au-Yong-Oliveira Effective and Potential Implications of Blockchain Technology for Auditing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435 Romildo Silva, Helena Inácio, and Rui Pedro Marques The Consequences of Key Audit Matters on Users: The Case of Spain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452 Tiago Ascenso and Cláudio Pais Enhancing Learning Object Repositories with Ontologies . . . . . . . . . . . 463 André Behr, Armando Mendes, José Cascalho, Luiz Rossi, Rosa Vicari, Paulo Trigo, Paulo Novo, Luis Cavique, and Hélia Guerra DevOps Certification in IT Industry: Preliminary Findings . . . . . . . . . . 473 Juanjo Pérez-Sánchez, Joaquín Nicolás Ros, and Juan Manuel Carrillo de Gea FGPE Gamification Service: A GraphQL Service to Gamify Online Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480 José Carlos Paiva, Alicja Haraszczuk, Ricardo Queirós, José Paulo Leal, Jakub Swacha, and Sokol Kosta Retraction Note to: Semantic Analysis on Twitter of Two Religious’ Leaders: @Pontifex and @DalaiLama . . . . . . . . . . . . . . . . . . . . . . . . . . Angel Torres-Toukoumidis, Isidro Marín-Gutiérrez, and Javier Herrán Gómez

C1

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491

Software and Systems Modeling

Analysis of German National Electricity Grid at Risk of Random Damage - Case Study Piotr Hadaj(B)

and Dominik Strzałka

Department of Complex Systems, Rzeszów University of Technology, Al. Powsta´nców Warszawy 12, 35-959 Rzeszów, Poland {piotr.hadaj,strzalka}@prz.edu.pl

Abstract. In this paper, a case study is conducted based on the real data obtained for German national electricity grid. The analyzed network represents connections and high-voltage switchgears of 220 and 380 kV lines. Selected graph parameters – the average vertex degree, vertex degree distribution, network efficiency, the clustering coefficient, the average path length, and the graph diameter were examined, taking into account that in the analysis, some nodes were randomly removed, simulating different failures. For each simulated failure, the effects on network parameters were calculated. As a final result, it was shown that in the analyzed case, the removal of only ten nodes could cause a significant fall of graph efficiency. In turn, this means that the whole considered network cannot be treated as a fault-tolerant one. Keywords: Complex networks · Network theory (graphs) · Systems interaction · Power industry · Power system security

1 Introduction The theory and practical applications of complex systems paradigm recently has shown many interesting approaches in real problems modelling [1]. There are many real systems that can be analyzed in terms of complex systems and having in mind that today without electrical energy there is no possibility to normally operate the importance of power systems their fault tolerance is crucial. They can be exposed on different randomly appeared threats finally even leading to the blackouts [2]. The paper shows a real electrical network (the case of Germany) as a net of vertices and nodes. Such a real network has own topological parameters that influence the power system security [3]. The idea of complex systems is supported by the concept of complex networks which have own topological parameters – their interpretation can be related to the stability of power grid and after the removal of some network vertices caused by the random failures the whole system even can collapse. The most important is the drop of network efficiency and in turn, the deterioration of quality parameters of supplied energy and increase of transmission costs [4]. The paper is organized as follows: after the short Introduction in Sect. 2 the theory of complex systems and networks is shown. The analysis of real source data based on © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 3–10, 2021. https://doi.org/10.1007/978-3-030-72654-6_1

4

P. Hadaj and D. Strzałka

NetworkX and Gephi software packages is given in Sect. 3. Section 4 shows the paper summary.

2 Complex Systems and Networks The concept of complex systems is an expansion of the term system S understood as set of beings B consisting of n–th element edge set E = {e1 , e2 , …, en }, which represent a system topology. They have m attributes represented by the set A = {a1 , a2 , …, am } and among them there are k possible long- or short-range relations given by set R = {r 1 , r 2 , …, r k }. This gives S = B(E, A, R) [3, 4]. In almost all cases the commonly accepted definition of complex systems is related to the Aristotle’s rule: the whole is more than the sum of its parts [5], however it should be noted that there are many other existing definitions based on the number of interdependent system elements, existence of different, sometimes even unknow dependencies that are long in time and spatial domains, system self-organization and emergence of new features. No matter which definition is taken into account, real complex systems have many interesting features. Some of them can be easily shown by the basic topological aspects expressed with graphs known as complex networks of nodes and relations (edges). Thus, in the simplest approach we have a graph G = {V, E} with n nodes (vertices V = {v1 , v2 , …, vn }) connected by m edges E = {e1 , e2 , …, em }. This graph can have some special features. For example, the sum of the ij all possible graph paths distances between the vertices i and j, ∀ i = j (here understood as the length of the shortest graph path d ij ) is usually very low even the network diameter is big. In paper [6] it was used to measure the scattering of the network. On the other hand, we can also define the network efficiency expressed as Eij = 1/d ij , ∀ i = j (if d ij = ∞, then Eij = 0). It is easy to conclude that networks with high efficiency have low costs of their operation; in our case for electrical networks, we can conclude that for such networks the costs of energy transmission is low. In the network vertices are connected to other ones. If the node i has k i edges that connect this vertex to k other nodes, one can check the ratio between the number E i of existing connections between these k i nodes and the number of all possible node edges having another interesting parameter, namely the clustering coefficient C i of node i. If in graph there is a clique, this parameter for each vertex I is equal to 1 (C i = 1), whereas for totally random graphs it goes to 0 (C i → 0). But the study [7] shows that in the case of complex networks this parameter can have different values leading to the concept of small worlds.

3 Case Study In this Section, the power grid data (based on the 220 kV and 380 kV connections) obtained for Germany (Fig. 1) has been entered into the yEd software package in the form of a graph. According to the report presented in [9] Germany had the biggest primary energy consumption in European Union in 2019 (in Europe the biggest energy consumption is in Russia). This is the biggest European economy and is strongly dependent on electrical energy. Some studies [10] have shown that this system is generally very stable and reliable if we refer to the international System Average Interruption

Analysis of German National Electricity Grid

5

Duration Index (SAIDI), but in Germany the grid operators have to go to great lengths to balance asymmetric production of green electricity. As a consequence, the problem of so-called “re-dispatch” has appeared. It is visible when the grid operator forces some power stations to lower the energy production because of a region oversupply, and other plants located in a low-production region are forced to increase the output. 0.29

0.29

0.29

0.14

Jardelund

Umspannwerk Lubmin 0.14

Lüdershagen

dania

Umspannwerk Lubmin

0.29

0.29

0.14

Bentwisch

FlenburgFlensburg

0.43

0.14

0.14

0.43

Bentwisch HGU

Kiel West Audorf Audorf 0.29

0.14 Inhausen

0.14

0.71

Hamburg Nord

Kiel-Süd

Umspannwerk Wilster

Konverstation Buttel

0.14

Umspannwerk Siedenbrünzow Umspannwerk Siedenbrunzow

0.14 0.43 Umspannwerk Lübeck

0.71 0.29

0.29 Voslapp

0.29

Hamburg Nord

Altentreptow/Nord 0.43

Umspannwerk Suderdonn

Güstrow (50Hertz) 0.71

0.14

0.14

0.29

Kummerfeld

Umspannwerk Kraftwerk Wilhelmshaven 0.43

0.43

0.14

Atlantic

Brokdorf

0.29

0.57

0.57

0.29

Itzehoe/West

0.29

0.29

Dollern

0.29

Dollern

Gos

Gustrow

Umspannwerk Hamburg/OST 0.29

Hamburg/Sud

Pasewalk

Umspannwerk Hernkraftwerk Brunsbuttel

Abbenfleth

Umspannwerk Maade

0.43

0.57

0.14 Umspannwerk Kernkraftwerk Brunsbüttel

0.29

0.29

Krummel

Altenreptow/Sud

DOW 0.29

0.57 0.14 Emden-Borssum

0.43

Unterweser 0.29

Conneforde 110/220 kV 0.29

Unterweser

Stade 0.29

Conneforde

Perleberg 0.29

0.29

Perlenberg

Alfstedt

0.43

0.29

Diele

Bertikow 0.14

0.57 0.43

Cloppenburg 220 kV

0.29

0.29

Farge

0.14 0.14

Lubeburg

0.71

Farge

Huntorf

Sottrum

0.14

Niedervieland

Umspannwerk Blockland

0.29

Ganderkesee

0.29

0.14

Sottrum

Vierraden

Dörpen West

0.43 Nahrstedt SUD 0.29 Wechold 0.29

0.29

Niederlangen

Stadorf

0.43

0.43

Umspannwerk Kraftwerk Robert Frank

Gransee

0.29 umspannwerk Kraftwerk Robert Frank 0.29 Meppen

0.71

0.14

0.29

0.14 Hannover/West

Umspannwerk Braunschweig Nord 0.29

Lahe TenneT TSO GmbH - Betriebszentrum Lehrte

Umspannwerk Hattorf

0.14 Station Sankt Hülfe 0.29 0.57 Umspannwerk Wolmirstedt

0.43 0.71

Umspannwerk Helmstedt

Umspannwerk Wolmirstedt

0.43 Hanekenfähr

0.43

Wahle

0.14

0.29

Wahle

Hanekenfähr

1.00

0.29

0.29

Landesbergen

0.29

0.14

0.43

0.29

0.43

Hennigsdorf

Umspannwerk Neuenhagen 0.29

Wustermark

Umspannwerk Malchow

0.43

0.43

Salzgitter-Hallendorf Neuehagen 0.29

Schaltanlage Ovenstädt

0.29

0.14 Buschhausen

0.43 Mehrum (220 kV)

0.43

0.14 Kraftwerkschaltanlage Ibbenbüren

0.14

Erzhausen

Umspannwerk Walzwerk (UWWW)

0.29

Station Westerkappeln 0.29

Station Ibbenbüren

Glindenberger Weg

Gleidingen 220 kV

Grohnde

0.43 Wehrendorf 0.43

Station Gronau Station Gronau

0.14

0.29

Godenau 0.29

Ohlensehlen

0.29 Wehrendorf

0.14 holandia

Lüstringen 0.14

Station Westerkappeln

0.29

KW Heyden

Heinersdorf4 0.29 Umspannwerk Thyrow 0.29

0.14

Station Hesseln

0.43

Umspannwerk Meißen

Algermissen 0.29

0.29

0.14

Godenau

Eickum 0.43

Brandenburg-West

Veltheim

0.29 Umspannwerk Bielefeld 0.43 Ost

0.57

Umspannwerk Bielefeld Ost Umspannwerk Bechterdissen 0.57

0.14

0.14

Blankenhagen

0.14

Umspannwerk Bechterdissen

Station Münster

Lage

0.29 Blankenhagen

0.29

0.29

0.29

0.29

Amelsbüren

Umspannwerk Förderstedt

0.29

Erzhausen

Station Roxel

0.29

Eisenhüttenstadt

Station Enniger 0.29

0.14

Station Gersteinwerk

Umspannwerk Paderborn Süd

0.71 0.43

0.29 Station Uentrop Umspannwerk Rosdorf

0.43

Station Gersteinwerk

0.14 Station Lippborg

0.29

Geithe

Umspannwerk Jessen Nord (Schweinitz)

0.14

0.29

Umspannwerk Schönewalde

holandia 0.29

0.29

0.29

Station Lippe

Station Nehden

Station Wambel

0.14

0.43 0.14

Umspannwerk Klostermansfeld

0.29

0.57 Umspannwerk Preilack

Station Unna 0.29

Station Kusenhorst

0.43

0.29 Station Schermbeck

0.57

Station Elmenhorst 0.57

Station Polsum 0.43

0.29

0.57

Station Arpe 0.29

Schaltanlage Datteln 4

0.57 Station Kusenhorst

0.43 Station 0.43 Niederrhein Station Niederrhein

Umspannwerk Ragow

Umspannwerk Marke 0.29

Umspannwerk Twistetal

Station Ratsbusch 0.29 0.43 0.14 Station Pfalzdorf

Umspannwerk Lauchstädt

Station Mengede

Station Polsum 0.29

Station Ossenberg

0.43

0.29

Station Bellendorf

Grohnde 380 kV / 110 kV

Station Pöppinghausen

0.29

0.29

UW Waldeck 0.14

Station Ellinghorst

0.43

0.43

0.43

Station Bergmannsglück 0.29

Graustein

0.29

0.14

Station Bochum 0.29

Station Möllen

Sandershausen

Station Kruckel

Station Nordlicht

0.29

kraftwerk

0.57

Station Zensenbusch 0.29

Station Hüllen

0.29

Witten

0.29

Bärwalde 0.29

Station Karnap

0.29

Station Hattingen Station Welper 0.29 0.29

Station Hattingen 0.43

Station Walsum

0.29

0.43

Streumen

Vörden Station Eiberg

0.29

0.43

Unterwerk Hagen

0.29

Station Utfort

0.29 0.43

0.14

Wolkramshausen (220 kV)

Taucha

Garenfeld

Station Utfort 0.29

0.57

0.29

0.14

WürgassenWürgassen

Station Ohligs

0.14

Station Genna

0.29 0.14

Umspannwerk Bixterheide

Station Edelstahl Krefeld 0.43

Station Dülken 0.14 Eula

Station Rheinhausen

0.14

0.14

Station Eller

0.14

0.29

Dillenburg

Station Rath 0.29

0.14 0.29

0.29

0.29

1.00

0.29

Station Dauersberg 0.43 0.43

Station Paffendorf

Station Knapsack

0.57

Umspannwerk Limburg

0.29

0.43 Station Oberzier

Station Siegburg

0.14

Umspannwerk Herlasgrün

0.71

Umspannwerk Vieselbach 0.57

0.43

0.29 Umspannwerk Karben Dörnigheim 220 kV

0.29

Station Alfter

0.14 Polska

0.29

0.14

0.14

0.43 0.29

0.86

Station Wengerohr

Umspannwerk Zwönitz 0.14 czechy

Umspannwerk Remptendorf 0.14

UW Dipperz

Umspannwerk Altenfeld

Markersbach

Umspannwerk Frankfurt Südwest

Station Sechtem

0.14

0.29

0.29

Station Ließem

0.14

Umspannanlage Zukunft

0.43

q 0.43

0.43

Umspannwerk Großkrotzenburg

0.14 Station Kelsterbach

0.29 0.29 Umspannwerk Aschaffenburg Karlstein

Station Weißenthurm

0.29

0.14

Pumpspeicherwerk Goldisthal Umspannwerk Großkrotzenburg 0.71

Umspannwerk Kriftel

Station Koblenz

Station Weißenthurm 0.43

Station Trier

Kraftwerkseinspeisung Staudinger 0.43

0.43

0.14

2 0.14

0.29 Hagenwerder

Umspannwerk Röhrsdorf

0.14 Umspannwerk Crossen

Umspannwerk Frankfurt Nord

Umformerwerk Köln

Station Sechtem

0.14

0.43

Dresden-Süd

Umspannwerk Weida

Mecklar

Umspannwerk Gießen Nord

0.43

0.14 Station Hambach

Umspannwerk Weida 0.43

Umspannwerk Aßlar

Station Dhünnaue 0.29

Station Dünnwald

0.29 0.43 Umspannanlage Brauweiler 0.29 Station Knapsack 0.14 Station Bollenacker

SA Bauler

0.14 Niederwiesa

0.29

0.43

0.29

0.14

UW Fühlingen

0.57 Umspannwerk Röhrsdorf 0.86

0.57

0.29KW Niederaußem

Station Verlautenheide

Eisenach

Umspannwerk Borken

Station Dauersberg

Setzer Wiese 0.29 Umspannwerk Bocklemünd 0.29

Umspannwerk Rommerskirchen Umspannanlage Brauweiler 0.43

0.29 Freiberg/Nord

0.29

0.57

Station 0.57Eiserfeld

Station Linde

Station Halfeshof

0.43 Station Opladen

0.57 Umspannwerk Rommerskirchen

0.29

Großschwabhausen 0.43

Station Halfeshof'

0.86 Station Sankt Peter 380 kV Station Opladen

0.29

0.29

Station Ronsdorf 0.29

Station Sankt Peter0.43 220 kV

0.14 holandia

Siersdorf

Umspannstation Lippendorf Block S

Umspannwerk Bergshausen

0.29 0.43

0.43 0.71 Station Gohrpunkt Station Gohrpunkt

Schmölln

Umspannwerk Pulgar

0.14 Umspannwerk Vieselbach0.71

0.29 0.14 Station Reisholz Station Norf

Station Osterath

0.14

KW Neurath 0.43 KW Neurath

0.43

Eula 0.43

Sandershausen (neu)

0.29 Elverlingsen

Station Mündelheim 0.29 Station 0.29Osterath

0.14 Umspannanlage Frimmersdorf I

0.29

0.14

Station Dahlem

0.29

1

0.14

0.43

fea 0.29

Umspannwerk Urberach 0.14

Umspannstation Bischofsheim

Umspannwerk Urberach

UA Merzig

0.57 0.29 0.29

0.29

Umspannwerk Redwitz

Umspannwerk Trennfeld 0.43

Umspannwerk Waldlaubersheim

Station Niederstedem 0.29

Umspannanlage Pfungstadt

Station Niederstedem

0.29

0.14 UA Obergraben

0.29 0.29

0.43 Dieffeln

0.29

0.57

0.14 0.71

0.29

Station Bürstadt

0.29

Station Osburg 0.14

Umspannwerk Bergrheinfeld/West

Umspannwerk 0.14 Mechlenreuth

0.43

Umspannwerk Bergrheinfeld

Biblis A

0.43

Station Quint

0.29

0.29 Umspannwerk Eltmann

Umspannwerk Mechlenreuth

Umspannwerk Würgau

Umspannwerk Oberhaid

Grafenrheinfeld

Beerfelden

Station Saarwellingen

UA Dillinger Hütte Mitte

0.14

0.29 0.29 0.29 Schwalbach Mühle

Umspannwerk Uchtelfangen 0.43

0.29

Station Rheinau 0.43 Mittelbexbach

0.43

Umspannwerk Kastenweiher

Umspannanlage Lambsheim

Umspannwerk Uchtelfangen

Umspannanlage Ensdorf 0.29 0.29 0.14

Umspannanlage Ensdorf

francja

0.14

0.29

Umspannwerk Kriegenbrunn

0.29

Schaltanlage St. Barbara

0.43

0.14

Umspannwerk Umspannwerk EtzenrichtEtzenricht

czechy

0.29 Umspannwerk Hüffenhardt

Umspannwerk Altlußheim

0.43

0.29

0.14

0.43

0.29 Umspannwerk Raitersaich Raitersaich Umspannwerk Umspannwerk Ludersheim

Umspannwerk Kupferzell

fra

0.29 Weingarten 0.57 0.29

0.29

SchwandorfSchwandorf

Umspannwerk Goldshöfe 0.14

0.29

Karlsruhe-West

EnBW-Umspannwerk Großgartach

0.43 0.29 Karlsruhe-Daxlanden Karlsruhe-Daxlanden 0.43 Schaltwerk Neckarwestheim

0.29 0.29 Ingolstadt

0.14

0.14

Umspannwerk Plattling

0.29 0.14 Umspannwerk Irsching Irsching Umspannwerk

0.29

Umspannwerk Rotensohl

Netze BW Umspannwerk Hoheneck

0.29

Ingolstadt

Umspannwerk Winnenden 0.29 Umspannwerk Mühlhausen

0.29 0.29

Umspannwerk Pulverdingen

220-kV-Umspannwerk Kuppenheim

0.43 0.29 0.29 Umspannwerk Oberjettingen

Umspannwerk Sittling

0.43

Zolling - Irsching

0.14

0.29

Umspannwerk Mühlhausen

0.29

Umspannwerk Niederstotzingen 0.14

Umspannwerk Neufinsing

Umspannwerk Bühl 0.14

0.43

0.14

Umspannwerk Oberbachern 0.29

0.14

Umspannwerk Ottenhofen

Station Meitingen 0.14

Gundremmingen

Hauptumspannwerk Menzing 0.29

0.14

Umspannwerk Marienberg

Föhring

0.29

Umspannwerk Lechhausen; Übernahmestation

0.14

austria

0.29

Umspannwerk Oberbachern

Schaltanlage Gundelfingen 0.43

Isar

Umspannwerk 0.57 Ottenhofen

0.43

0.57 0.14

Netze BW0.57 Umspannwerk Bünzwangen Umspannwerk Wendlingen 0.29

Umspannwerk Trossingen

0.29 0.29

Umspannwerk Meitingen

Netze BW Umspannwerk Endersbach

0.29 Umspannwerk Engstlatt

0.43 Umspannwerk Weier

0.29

Umspannwerk Pleinting 0.29 Umspannwerk Pleinting Umspannwerk Simbach 0.29 Umspannwerk Pirach Umspannwerk Altheim

0.29

Umspannwerk Oberbrunn

0.29

Umspannwerk Villingen

Netze BW Umspannwerk Laichingen

0.43

0.29 380-kV-Umspannwerk Eichstetten 0.29

0.71

0.43

Vöhringen LEW 0.57

0.29

220-kV-Umspannwerk Eichstetten

0.29

Umspannwerk Oberottmarshausen Umspannwerk Metzingen

Umspannwerk Dellmensingen

Vöhringen Amprion

Umspannwerk Krün

0.29

Umspannwerk Gurtweil 0.14 0.43Tiengen

Lastverteilung/Schaltanlage Kühmoos 0.14 Schwörstadt 0.14

Umspannwerk Woringen

0.14

0.29 0.71 Lastverteilung/Schaltanlage Kühmoos

0.14 francja

0.14 0.57 Laufenburg

Tiengen

0.29

0.43

Umspannwerk Bidingen Nr. 697 S

Umspannwerk Herbertingen 0.43 Umspannwerk Herbertingen

Laufenburg

Asphard 0.14 szwajcaria

0.43

0.29

380kV-Umspannwerk Grünkraut

Umspannanlage Leupolz

0.14 aus

0.14 austria 0.14 austria

Fig. 1. Original graph based on German electricity grid with centrality analysis (yEd software).

In our approach each network node in Fig. 1 represents a separate vertex of the graph. The connections between nodes were recreated as graph edges, which means that the network of graph connections is the same as the real layout of the transmission network. The initial form of created graph has 452 vertices and 490 edges. The average vertex degree of the original graph is 2.168, and the calculated average efficiency of the graph is E(G) = 0.042. The average path length in the graph is  = 11.996, and the diameter is 30. The vertices degree parameter was calculated basing on the number of edges connected to each node. As energy flow can be bidirectional, edges haven’t got the direction (are undirected).

6

P. Hadaj and D. Strzałka

Usually graphs representing real power grid systems can be analyzed in many ways, presented approach was also used in [11]. For example, in Fig. 1 the analysis of graph centrality was performed (basing on the approach proposed in [8]): the bigger the graphic size of the node is the more connections to it exists – the vertex degree in graph is higher. Moreover, some rings of a large diameter network can be distinguished, which correspond to the actual connections in the electricity grid. The loops and bypasses are a key to increase the transmission network reliability. In case of a failure of a low-degree node, the ability to transmit electricity is maintained. The proposed simulation process was as follows: a random node was selected and removed from the network. Then the graph parameters were analyzed. Then another random node was removed and the analysis was carried out. It was repeated until 10 vertices were removed. The described process was repeated at least 10 times to make the results credible and average. The scheme and the order of the nodes elimination are shown in Fig. 2, while the averaged values from the last stage, along with the rest of the data, are described in Table 1. Figure 4 illustrates the change of graph vertices degree distribution between original network state and the final stage of simulation. The value of k = 2 indicates that many vertices belong to different chains (sometimes rings). When some of them are removed, as a consequence some parts of the network (that do not belong to any ring) can be easily separated.

Fig. 2. Simulation procedure.

The analysis of the simulation results showed a significant decrease in the value of the network performance parameter and a significant increase in the average path length in the analyzed transmission structure. Figure 3 illustrates changes of mentioned graph parameters through the simulation procedure.

Analysis of German National Electricity Grid

7

14.5

0.044 0.042

14

0.04

13.5

0.038

13

0.036 0.034

y = -0.0008x + 0.042 R² = 0.9757

0.032

y = 0.1733x + 12.242 R² = 0.7222

12.5 12 11.5

0.03 0

1

2

3

4 5 6 7 8 Number of removed nodes

9

10

11

0

1

2

3

4 5 6 7 8 Number of removed nodes

9

10

11

Fig. 3. Changes of the network efficiency (left) and average network path length (right) in successive steps of simulation.

Some of graph vertices and groups of vertices were separated from its main part (see Fig. 4 and consider k = 2); however, it can be seen that a significant proportion of the nodes are sufficiently connected to the network; the transport of energy can be ensured.

200

Number of verƟces

150

100

Original network Final stage

50

0 0

1

2

3

4

5

6

7

Vertex degree

Fig. 4. Comparison of vertex degree distribution of original network state and after the final stage of simulation.

The meaning of the parameters in the Table 1 is described below [12]:

452

490

2.168

0.02905

0.04212

0.039

11.996

30

Nodes in graph

Edges in graph

Average vertex degree

Local efficiency

Average efficiency

Average clustering coefficient

Average path length

Graph diameter

0

Number of removed nodes

31.23

12.071

0.04

0.04178

0.02912

2.146

485.5

451

1

33.14

12.868

0.04

0.04036

0.02918

2.129

479.33

450

2

33.89

12.921

0.041

0.03914

0.02925

2.107

473.5

449

3

34.05

12.922

0.041

0.03851

0.02931

2.094

469.66

448

4

34.32

12.89388

0.042

0.03782

0.02938

2.081

465.12

447

5

Table 1. Simulation results.

38.54

14.1821

0.042

0.0362

0.02945

2.063

460.16

446

6

38.74

13.5084

0.042

0.03563

0.02951

2.054

456.2

445

7

38.79

13.5508

0.042

0.03517

0.02958

2.045

454.16

444

8

38.91

13.5621

0.043

0.03523

0.02964

2.036

451.5

443

9

38.98

13.7192

0.043

0.03365

0.02971

2.014

445.33

442

10

8 P. Hadaj and D. Strzałka

Analysis of German National Electricity Grid

9

The number of edges and nodes represents the physical number of objects: transmission lines are edges, transformers and power distribution stations are nodes. The average vertex degree is the average number of connections (edges, 220 kV and 380 kV power lines) per node – does not have to be high for network security, as opposed to efficiency and clustering coefficient. The average clustering coefficient parameter for node i was defined in Sect. 2 and for the whole graph this parameter is calculated as an average of the overall graph nodes. The local efficiency (calculated per node) and the average efficiency are the parameters that describe the considered system efficiency according to the details given in Sect. 2. The average path length and graph diameter describe the physical properties of the system. Its value results from averaging the length of the shortest paths for all pairs of vertices. Small average path length means that the graph has a compact structure, allowing for effective flow of information in graph. Graph diameter represents the highest of the shortest path between each pair of vertices in the graph. If these parameters have high values the system is geographically extensive and our simulations of the node failures reduce them because of the divisions of the whole network into independent and smaller parts. In turn they are not able to ensure energy transport.

4 Summary This paper shows the analysis of some complex network parameters related to the real data of electrical grid network. The randomized removal of 10 nodes in analyzed network caused the drop of network average efficiency by 25%. The efficiency is related to the cost of whole network operation: high efficiency means low costs of energy transmission in the network. In the case of the breakdown, self-failure or accidents among the nodes they can significantly increase the cost of whole network operation. We show that in the case of German national electricity grid it is enough to break 10 vertices to cause such a situation. This could be in the contradiction to the study given [10] where basing on SADI index the system shows stability and reliability. However, we show that the breakdown of small number of nodes in 220 kV and 380 kV grid network can cause a very serious consequences. In the feature work, it is assumed to analyze the removal of more than 10 nodes and this will be repeated at least 100 times. It is also possible to relate this approach to other countries; however, it should be remembered that only for countries that have sufficient number of nodes, the obtained results will be reliable. Because such national networks are interconnected, the research can be even carried on data related to the whole EU or Europe.

References 1. Newman, M.; Networks: An Introduction, Oxford University Press, Inc., New York (2010) 2. Wu, L., Anderson, R.N., Boulanger, A., Rudin, C., Kaiser, G.E.: Failure Analysis of the New York City Power Grid; CU CS Technical Report CUCS-025-14; Department of Computer Science. Columbia University, Columbia (2014)

10

P. Hadaj and D. Strzałka

3. Sachtjen, M.L., Carreras, B.A., Lynch, V.E.: Disturbances in a power transmission system. Phys. Rev. E., 61(5), 4877–4882 (2000) 4. Łatka, M., Nowak, M.: Analysis of electrical power quality parameters at the high voltage level. In Proceedings of the 2018 Conference on Progress in Applied Electrical Engineering (PAEE), Ko´scielisko, Poland, 18–22th June 2018 (2018) 5. Upton, J., Janeka, I., Ferraro, N.: The whole is more than the sum of its parts: aristotle, metaphysical. J Craniofacial Surg. 25, 59–63 (2014) 6. Holmgren, Å.J.: Using graph models to analyze the vulnerability of electric power networks. Risk Anal. Int. J. 26, 955–969 (2006) 7. Latora, V., Marchiori, M.: Efficient behavior of small-world networks. Phys. Rev. Lett. 87, (2001) 8. Hines, P., Blumsack, S., A centrality Measure for Electrical Networks, IEEE 41st Hawaii International conference on system sciences, 2008 9. Sönnichsen, N.: Primary energy consumption in selected countries in Europe in 2019 (2020). https://www.statista.com/statistics/332520/primary-energy-consumption-in-sel ected-countries-in-europe/ 10. Amelung S.; Schlandt J.: Germany’s electricity grid stable amid energy transition (2018). https://reneweconomy.com.au/germanys-electricity-grid-stable-amid-energy-tra nsition-13244/ 11. Hadaj, P., Strzałka, D.: Modelling selected parameters of power grid network in the southeastern part of poland: the case study. Energies 13(1), 239 (2020) 12. Crucitti, P., Latora, V., Marchiori, M.: A topological analysis of the italian electric power grid, Physica A: statistical mechanics and its applications. In: 2nd International Conference on Frontier Science, Proceedings of the Conference A Nonlinear World: The Real World, vol. 338, no. 1-2, pp. 92–97 (2004)

Advanced Smart Car Park Management System for High Car Park Utilization Ammar Sami Aldallal(B) , Aysha Al Zallaqi, and Amina Mohammed Computer Engineering Department, Ahlia University, Manama, Bahrain [email protected]

Abstract. Finding a car park in a crowded parking lot is becoming a challenge that is attracting increasing interest from both industrial and academic researchers. This paper aimed to provide an efficient utilization of a parking lot by solving the problem of wrong parking without using monitoring system or human intervention. This is accomplished through developing a multi-functional smart car parking system that is more cost effective and user friendly than existing systems. The proposed system utilizes a number of low-cost light dependent resistors and IR sensors connected to a raspberry pi as a control unit. Simple to use web application is also developed to receive the data from the raspberry pi that will display the location of empty parking slots on user’s smart phone. The project was highly successful attaining 100% utilization of parking slots, no congestion occurred simply by preventing cars from entering the park if no parking spaces were available. No car can occupy more than one slot. Locations of empty slots are accurately displayed on the web application; hence, facilitating the process of reaching the empty slot quickly. This system is beneficial for drivers as it will enable them to locate the exact empty parking slot easily, so they save their time and consume less fuel chasing an empty slot. In addition, no congestion will occur inside the parking area. Moreover, this system is also beneficial for car park owners as the utilization of the car park become more efficient where the number of parking cars equals to the number of parking slots. Keywords: Smart car park · Car park · Raspberry pi · Parking lot · Park utilization · IR sensor

1 Introduction The increasing number of car ownership globally has resulted in added stress for drivers. In fact, this research was driven by the difficulty car owners face to locate a parking slot for a car due to the increasing number of car users. In certain cases, such as downtown areas it is found that it takes the driver an average of 20 min to find a parking slot as per a global parking survey by IBM [1]. On average, it is found that 30% of traffic congestion is due to chasing an empty car park [2]. This results a great waste of drivers’ time and loss in fuel efficiency while searching for an empty parking slot. In 2006, for example, a study in France showed that 700 million Euros is the annual loss suffered for searching for a park [3]. In a more recent study, the congestion time loss was estimated to be $1433 © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 11–21, 2021. https://doi.org/10.1007/978-3-030-72654-6_2

12

A. S. Aldallal et al.

annually per auto commuter in Washington [4]. In some cities such as business distinct in Los Angeles, annually, cars searching for a park burn 47000 gallons of gasoline, produce 730 tons of carbon and create the equivalent of 38 trips around the world [5]. By looking at the causes of parking problems, it is found that they are based on two major problems. The first one is the unorganized parking, in which drivers do not park accurately between lines resulting in wrong parking where some drivers occupy two slots by one car either for their lack of experience or any other reason. This problem is described as inefficient use of existing park capacity [6] or car park utilization [7]. The other problem is entering the parking lot although there is no empty parking slot which leads to congestions due to accumulation of parking seekers. Researchers tried to improve the utilization of car park using different methods. The most common method is through image processing of the car par area [8]. [7] determined the occupancy states of the car park based on the visual feature s extracted from parking slots by using trained neural networks. Researchers were inspired to create smart parking systems to solve these issues. In this research, a prototype for multi-function car park management system (MFCPMS) is developed to provide drivers with many features that help them to easily locate a free slot to park in. This system is more efficient in buildings, malls, universities, airports, hotels, hospitals, and any other free-car parking lot. The rest of this paper is organized as follows: Sect. 2 reviews the related work. In Sect. 3 methodology and design of the proposed system including the circuit diagrams, block diagrams and flow chart diagrams are presented. In Sect. 4 integration of the components and the functionality is presented. Results are analyzed in Sect. 5. Finally, Sect. 6 provides the conclusion and further enhancements.

2 Related Work The car park problem is a worldwide issue because the number of vehicles is increasing rapidly. Drivers believe that it is very hard to find an empty park as they would like especially close to the theaters, airports and shopping malls. The lack of accessible parking affects local businesses and visitor’s quality of life. The following list identifies a number of problems that typically occur in many communities: • Not enough information for drivers on parking availability and price. Drivers find it frustrating when they are expecting free spaces to park in but in reality they face a limited number of spaces and expensive parking. • Wrong park situation. It happens when a driver takes more than one parking slot for his or her car. This will affect the availability of the parking space and will make it harder for others to find a free slot. Entering the parking area although there is no free slot. It happens when there is no indicator at the entrance to tell the drivers that the parking lot is full. By allowing cars to enter in this situation causes congestion inside parking area. In order to overcome these issues, researchers worked hard to come up with different techniques and solutions.

Advanced Smart Car Park Management System for High Car Park Utilization

13

Al-Kharusi and Al-Bahadly [9] developed an intelligent car park system based on image processing. A camera is used to capture an image of full parking lot when it is empty. Then several captures are taken on different timing slot that are then processed through several image processing stages in order to identify the location of the empty slots. This system is composed of hardware and software. The hardware composed of video camera that takes continuous images of the parking lot. These images are then transmitted to the PC through radio transmission to be processed by MATLAB software to guide the drivers to the empty parking slots. The main aim if this work was t identify the location of empty slots. And no results about the percentage of utilization are presented. Results show better performance of the proposed system. [10] developed a simulation of smart parking system using MATLAB. It is based on secured wireless network and sensor communication. Through this system drivers can view and reserve parking slots, receive advertisements of discount of parking fees once they are passing by and the system can predict the revenue of the parking site. The results are analyzed in terms of average searching time of empty slot and average occupation percentage compared to the traditional parking system where drivers search by themselves for empty slots. [7] Developed a vacancy detection system based on image processing of a 24-h video captured for the parking area. The occupancy states of the car park is determined based on the visual features extracted from parking slots by using trained neural networks. This method is powerful in responding to the light intensity changes. Also it is capable of detecting vacancies in nighttime. This method achieves accuracy of identifying the empty slots by 99.9%. An iParker system that offers multiple features is proposed by Kotb et al. [8]. These features include park reservation, reduced cost and reduced searching time. This system is based on a mathematical modeling that uses mixed-integer linear Programming. Reservation of the parking slot can be either real time reservation (dynamic reservation) at which the parker can spend as much time as he wants or shared time (static reservation) where the parker is given a specicf slot of time to utilize for parking. Several analyses are performed to examine its performance. The results show that iParker system can achieve up to 28% cut in the total effective cost for all parks and the total utilization increase by 21% compared to non-guided parking system. However, if the parker decides to use static reservation and leaves earlier than the expected time slot, then the utilization of parking slot will decrease. As cars have different sizes, the parking space depends on their size. From another prospective, some drivers do not park their cars within the designate area, causing occupation of more than one slot, resulting in a partially free parking slot. Samaras et al. [11] simulated a parking system using MATLAB to measure the impact of web services on WSN when they are used to estimate the free size of parking spaces. Although the concept is worthwhile the main concern of the study was on the energy consumption of the battery of the WSNs. Moreover, this system can be implemented in an unorganized parking area where the paring area depends on the parked car size. In order to reduce the time required to allocate an empty parking slot and to avoid congestion inside the parking area, several approaches are developed. Geng et al. [12] proposed a smart parking system where an optimal parking space is assigned and reserved

14

A. S. Aldallal et al.

to a driver based on his location. At each defined decision point in a time-driven sequence the system evaluates a mixed-integer linear programming value which produces an optimal allocation according to his location. Compared to the state-of-the-art guidance-based systems, the average required time to find a parking lot and the parking cost are reduced. Hanche et al. [13] proposed an automated parking system using RFID. In their system, the entrance and exit will be fast since each driver will have an RFID tag that will be used to process the payment of parking cost without stopping at the circulation point. This method reduces the congestion at the inlet and exit of the parking slot. Based on [14] an optical wireless sensor network (WSN) was developed for traffic monitoring which can be applied to monitor vehicles in a parking area. The system can inform drivers of the available parking spaces then directs them to it. This kind of system would make it very easy for drivers to find a free slot in a crowded parking garage. Yang et al. [15] developed a prototype of smart parking system consists of set of light wireless sensor networks mounted for each parking slot, connected to embedded webserver to display in a real time the location of the empty slot on a Google map. However, this system is not applicable for in-building parking slots as they cannot appear on the Google map. In addition to locating an empty parking slot in a parking area, there is a need to have a monitoring system that can identify people who cause damage to other cars or blockage of other cars due to incorrect parking. The damage could be hitting, denting, scratching or scraping the car. Aalsalem et al. [16] proposed a parking management system that monitors the car park through Automatic Number Plate Recognition Cameras (ANPR) connected through LAN to the DB that has the information of the car owner associated to the car license plate number. This system is applicable for parking area that is utilized by pre-defined drivers such as university parking area. It is noted from the literature that researchers developed several approaches to solve the problem of parking utilization. Mostly they focus on software solutions either through image processing to identify empty locations or through mathematical algorithms to locate and reserve a parking slot. In this work we are going to tackle the problem of parking utilization by using hybrid solution. This solution composed of both software and hardware components. This work aims to solve the aforementioned three problems in one system, making it more cost effective and reliable for both car park owners and car drivers. What motivated proponents to undertake this project was based on these technologies: low cost Light Dependent Resistors (LDR) controlled by raspberry pi to provide highly utilized car park and friendly web application-based system to guide the user to the exact location of empty parking slot.

3 System Overview This section includes description of the main components utilized in this research along with their features and functionalities. Although some of them are common in designing a wireless network system, there is a need to emphasize the advantages of using these components. The following is a list of the main hardware and software components used to build the proposed system highlighting their feasibility from the viewpoint of

Advanced Smart Car Park Management System for High Car Park Utilization

15

the mutli-functional smart car parking system. These components are categorized into hardware and software components. 3.1 Hardware Components The following section will describe the main components along with the rationale behind adopting them in this project. 1) Raspberry Pi It is a low cost tiny computer which looks like a credit card and it can be connected to a monitor, mouse and a keyboard. It can be explored and programmed by computers. Some of its applications are playing high definition videos, browsing the internet and playing games. The following are some of its characteristics: It contains 1 GB SDRAM, interface for the camera, interface for the display, 40 general purpose input and output pins, CPU 900 MHz quad-core ARM Cortex-A7, Micro SD card slot, 3.5 mm phone jack audio output, 4 USB 2.0 ports, HDMI video outputs, and operated by Raspbian: Debian Based Computer Operating System for Raspberry Pi. 2) Light Dependent Resistor (LDR) The second component is the light dependent resistor (LDR) which is a resistive device whose resistivity is a result of the incident electromagnetic radiation. Therefore, it is also called photocell, photoconductor or photo conductive cell. High resistance semiconducting materials are used for making these LDRs. Different symbols are used for LDR indication. The photoconductivity, an optical phenomenon, is the mean by which the conductivity of the materials is increased by light absorption. The electrons are excited to the condition band of the semiconductor material when light falls on the device. The energy possessed by the photons in the incident light should be greater than the semiconducting material’s band gap in order to make the electrons bounce to the conduction band from the valence band. When the device is stroked by light of enough energy, electrons are attracted to the conduction band which produces large number of charge carriers. At this stage the circuit closes so the current starts flowing through the device and the resistance of the device decreases as a result of this process. 3) IR Sensor An infrared sensor is a device that can recognize specific characteristics from its surroundings, it is done either by emitting or by detecting infrared radiation. It can also detect motion and measure the amount of heat that is being emitted by an object. The IR has two parts: receiver and transmitter. The transmitter sends an unseen beam to the receiver. If there is nothing that crosses this beam the sensor will keep on sending signals, but when any object crosses this beam the signal will get back to the transmitter and it will detect that there is an object that prevents the signal. This sensor has been used in this project to be placed on the borders between parking lots and it is responsible for detecting a wrongly parked car, i.e. parking on the line separating two parking slots. 4) DC motor All the electro-mechanical movements seen nowadays are caused by either an AC or a DC motor. Here the DC motor is used. It’s a device that converts DC electrical

16

A. S. Aldallal et al.

energy into a mechanical energy. The principle of working a DC motor is that it experiences a mechanical force whenever a current carrying a conductor is placed in a magnetic field. Fleming’s left-hand rule is used to give the direction of the force, the magnitude is given by F = B I L, where B is the magnetic flux destiny, I is the current and L is length of the conductor in the magnetic field. 3.2 Software Components One of the objectives of this project is to have an easy to use and friendly user interface. To accomplish this several software components were used to integrate and communicate with the hardware components. The first component is the main programming language which is python. It is a general dynamic programming language that allows programmer to code the concepts of programming using less number of lines of codes compared to C++ and Java. The second component is the Cascading style sheets (CSS). It is used for defining the style of the webpage, including the design, layout and variations of displaying different devices and screen sizes. It also describes how HTML elements are displayed on the screen. CSS saves a lot of programming work because it controls the layout of multiple web-pages at once. In addition to these components, Java script and Google Cloud Platform are utilized in implementing the proposed MCPMS.

4 Prototype Architecture This section explains the integration and implementation of the hardware and software components. It describes the main hardware circuits, flowchart and working principle of these components. 4.1 LDR Circuit Diagram One of the major components of the circuit which is the Light Dependent Resistor (LDR). In this project there was 6 LDRs used for the parking area, one for each of the six parking slots and one at the entrance of the parking lot. It has a high resistance in a bright place and a low resistance in a dark place, it was used in this project to detect whether there was a free slot or no. When a car passes over the LDR sensor it will automatically detect that there is a car placed in the parking area and then it will send this information to the raspberry pi. 4.2 Detecting Wrong Parking IR receiver and transmitter of the sensor circuit is used to perform this task. The transmitter sends a beam to the receiver. If nothing crosses this beam the sensor will keep on sending signals, but when any object crosses this beam the signal will be reflected to the transmitter and it will detect that there is an object that crosses the signal. This sensor is used in this project to be placed on the lines separating parking slots and it is responsible to detect the car when it parks wrongly, i.e. parking on the line separating two parking slots.

Advanced Smart Car Park Management System for High Car Park Utilization

17

4.3 Flow Chart Diagram The flow chart diagram of the project is illustrated in Fig. 1. The process starts by retrieving the gate LDR status and checking whether the park is full or not. If it’s full the gate will not open. If the park is not full the gate will open allowing the car to enter. Once the car takes place in a free slot, the corresponding LDR sensor will detect the object due to the change of resistance that is altered when light intensity falls on it. As a result, LDR status will be set to one. This value is communicated to the raspberry pi which in turn will pass it to the corresponding method that controls the Web application. The website application will be updated with the parking slot position marked as occupied by displaying a car image in the same place that the driver parked in. The website will also show the number of the available parks. Another thing to be checked upon parking is whether IR beam has been crossed during parking or not. The IR was mounted on the border line between parking slots. If the IR beam is crossed for more than 5 s the alarm will be switched on. The alarm will be switched off only when the beam is no longer crossed. Finally, the system will check the number of cars in the parking lot. If the park is full a red LED light will be switched on, otherwise a green LED will be switched on. As an added value feature the web application will allow users to have the option to add their feedback in the website. Sample of partially full parks as they appear in the web application is shown in Figs. 2.

Fig. 1. Flow chart of the proposed system

18

A. S. Aldallal et al.

Fig. 2. Partially occupied park as displayed in the web application

5 Results and Discussion The proposed car park management system was simulated using a prototype consisting of 6 parking slots. Initially the gate was closed. In front of the gate there was an LDR sensor which was used to sense the presence of new entering cars. By checking the status of availability of slot, the gate opens. Once the car takes its position in the park, the web application displayed on the smart phone was updated successfully with the exact slot shown as occupied (Fig. 3). This process continued until the car park was full. In this case the counter of occupied slots is equal to the total number of parking slots. Once the parking lot was full, the red LED was set and green LED was switched off. The gate won’t open for incoming cars. Once car leaves the parking lot through the exit gate, the counter of empty slots will be incremented, allowing for new cars to enter the parking area.

Fig. 3. Simulation for half full parking lot. The green LED is ON indicating availability of empty slot(s). In the back is shown the locations of empty and occupied slots as they appear on the Web application on the mobile phone

If the car parks wrongly (Fig. 4), it crosses the IR beam causing the buzzer to switch on alerting the driver to correct his parking position to be between lines. This project achieved all the objectives stated at the beginning. The car park was efficiently utilized. The number of parked cars equaled the number of parking slots which

Advanced Smart Car Park Management System for High Car Park Utilization

19

Fig. 4. Simulation of wrong parking. This case causes the Buzzer to alert the driver about wrong parking

implied 100% utilization. The empty parks were instantly and accurately indicated in the web application; hence saving the driver’s time looking for empty location. No cars were allowed in if the car park was full which eliminated the congestion inside the parking area. Although many researchers have developed car parking management systems, very few have considered wrong parking problem such as Wiseman [17]. However, his paper does not specify how this wrong parking avoidance can be implemented, so the suggested model of this paper can be also used to avoid wrong parking of autonomous cars. The proposed system paid a great attention to this problem. Not only by detecting wrong parking but also preventing it through alerting system that forces the driver not to occupy more than one slot. This feature works instantly. It eliminates the use of monitoring cameras and human interaction. With this feature implemented successfully, the car park utilization reaches its maximum of 100%. In addition, it avoid loss of income of paid parks, the case that could happen when incorrect parking causes the other park not be paid for by the wrongly parked car.

6 Conclusion In this research a multi-function car park system was developed. It utilized low cost LDR sensors fixed at each parking slot and controlled by a raspberry pi. These were used to identify the empty/occupied parking slots status. IR sensors were used to alert the driver in case of wrong parking between two slots. Users were guided to the exact empty slot by the developed simple-to-use web application via their smart phone. A prototype consists of six parking slots was built to simulate the functionality along with associated website to display the location of empty/occupied slots on the smart phone. This project resulted in high car park utilization attaining 100% utilization of parking slots, no congestion occurred due to to prevention of entering cars to the park if no slots were available. No car could occupy more than one slot. Locations of empty slots are accurately displayed on the web application; hence, facilitating the process of reaching the empty slot quickly. With this system the owners of car parking areas are satisfied

20

A. S. Aldallal et al.

that no single parking slot is wasted by wrong parking. Hence they gain the maximum possible income. Normally, in such situation, some keen car parking owners hire or assign a number of employees just to check wrong parked cars and flag them for extra charge. Further improvements to this system can be achieved by linking many parking areas into one application where the driver can select the parking area s/he is aiming to park in. This system can be integrated with other available systems to allow the driver to reserve a parking slot after identifying a suitable available location.

References 1. IBM Global Parking Survey: Drivers Share Worldwide Parking Woes, 2011, IBM, Armonk, NY, USA, September 28. https://www-03.ibm.com/press/us/en/pressrelease/355 15.wss. Accessed 18 Jan 2019 2. Shoup, D.C.: Cruising for parking. Transp. Policy 13(6), 479–486 (2006) 3. Gantelet, E., Lefauconnier, A.: The time looking for a parking space: strategies, associated nuisances and stakes of parking management in France. In: Proceedings ETC, pp. 1–7 (2006) 4. Bajpai, J.N.: Emerging vehicle technologies & the search for urban mobility solutions, Urban, Planning and Transport Research, vol. 4(1) (2016). https://doi.org/10.1080/21650020.2016. 1185964 5. Hashimoto, S., Kanamori, R., Ito, T.: Auction-based parking reservation system with electricity trading. In: Proceedings IEEE 15th CBI, July, pp. 33–40 (2013) 6. Rico, J., Sancho, J., Cendón, B., Camus, M.: Parking easier by using context information of a smart city: enabling fast search and management of parking resources. In: 2013 27th International Conference on Advanced Information Networking and Applications Workshops, Barcelona, pp. 1380–1385 (2013) 7. Jermsurawong, J., Ahsan, M., Haidar, A., Dong, H., Mavridis, N.: Car parking vacancy detection and its application in 24-hour statistical analysis. In: 10th International Conference on Frontiers of Information Technology, Islamabad, 17–19 December 2012, pp. 84–90 (2012) 8. Kotb, A.O., Shen, Y-H., Zhu, X., Huang, Y.: iParker-a new smart car-parking system based on dynamic resource allocation and pricing. IEEE Trans. Intell. Transport. Syst. 17(9), 2637– 2647 (2016) 9. Al-Kharusi, H., Al-Bahadly, I.: Intelligent parking management system based on image processing. World J. Eng. Technol. 2, 55–67 (2014) 10. Yan, G., Yang, W., Rawat, D.B., Olariu, S.: SmartParking: a secure and intelligent parking system. IEEE Intell. Transp. Syst. Mag. 3(1), 18–30 (2011) 11. Samaras, I., Gialelis, J., Koubias, S.: Using web services-based wireless sensor networks for estimating the free size of parking places. In: 2016 23rd International Conference on Telecommunications (ICT), Thessaloniki, pp. 1–5 (2016). https://doi.org/10.1109/ict.2016. 7500439 12. Geng, Y., Cassandras, C.G.: New “Smart Parking” system based on resource allocation and reservations. IEEE Trans. Intell. Transport. Syst. 14(3), 1129–1139 (2013) 13. Hanche, S.C., Munot, P., Bagal, P., Sonawane, K., Pise, P.: Automated vehicle parking system using RFID. ITSI Transactions on Electrical and Electronics Engineering (ITSI-TEEE), 1(2), 88–91 (2013) 14. Chinrungrueng, J., Sunantachaikul, U., Triamlumlerd, S.: Smart parking: an application of optical wireless sensor network. In: Proceedings of the International Symposium on Applications and the Internet Workshops (SAINTW 2007), pp. 66–69 (2007)

Advanced Smart Car Park Management System for High Car Park Utilization

21

15. Yang, J., Portilla, J., Riesgo, T.: Smart parking service based on Wireless Sensor Networks. IECON 2012 - 38th Annual Conference on IEEE Industrial Electronics Society, Montreal, QC, 2012, pp. 6029–6034 (2012) 16. Aalsalem, M.Y., Khan, W., Z., Dhabbah, K.M.: An automated vehicle parking monitoring and management system using ANPR cameras. In: 17th International Conference on Advanced Communication Technology (ICACT), Seoul, 2015, pp. 706–710 (2015). https://doi.org/10. 1109/icact.2015.7224887 17. Wiseman, Y.: Remote parking for autonomous vehicles. Int. J. Hybrid Inf. Technol. 10(1), 313–322 (2017)

Analyzing and Modeling Critical Risks in Software Development Projects: A Study Based on RFMEA and Systems Dynamics Lucilene da Silva Leite, Alderedo Rubem de Farias Neto(B) , Fernanda Lopes de Lima, and Ricardo Matos Chaim Department of Computer Science, Universidade de Brasília, ICC Centro - Módulo 14, Subsolo CSS-361 - Campus Darcy Ribeiro, Brasília, DF, Brazil [email protected], [email protected], [email protected], [email protected]

Abstract. This article aims to propose a method of qualitative analysis and risk management in software development project environments, using a combination of RFMEA and System Dynamics. Therefore, it was possible to analyze the main risks in software development projects from the literature review. Based on those risks, a survey was applied with experts of one of the projects of the Brazilian Civil Aviation Agency (ANAC) to estimate the probability of occurrence, impact, and detection of each risk. Through RFMEA, it was possible to identify and prioritize the project’s risks, and with Systems Dynamics, to evaluate the causal relationships among them. This is a case study and qualitative research carried out with the validation of ten risks in a total of 23, establishing the strategies and recommended actions for each one of them. Keywords: Risk management · Software development project management · RFMEA · System dynamics

1 Context According to ISO 31000 - Risk Management - Guidelines (ISO 2018), the risk is the effect of uncertainties on objectives. The same standard also defines risk management as a set of coordinated activities to direct and control an organization regarding risks and create and protect value. The standard further explains that risks can emerge, change, or disappear as an organization’s external and internal contexts change. To that end, risk management anticipates, detects, recognizes, and responds to these changes and events in an appropriate and timely manner. Based on Carbone and Tippet (2004), effective risk management is essential in project management. Proper risk management can help the project manager mitigate known and unforeseen risks in all projects. The absence of effective risk management can cause projects to over budget, schedule delays, meet critical performance goals, or present another combination of these problems. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 22–35, 2021. https://doi.org/10.1007/978-3-030-72654-6_3

Analyzing and Modeling Critical Risks in Software Development Projects

23

Certainly, several factors determine whether a project will be successful, but failing to carry out risk management will increase the possibility of failures. As a result, it is essential having an effective method of planning and managing project risks, and that at the same time is easy to apply and disseminate with a team. Considering the above information, this article aims to propose a method for analyzing and managing risk in software development project environments, using a combination of two methods, RFMEA and Systems Dynamics. The analysis of critical risks will be carried out based on a simple tool, which provides benefits in the identification and treatment of risks related to projects, RFMEA (Failure Mode and Effect Analysis), associated with a widespread method in risk management: Systems Dynamics. Systems Dynamics will be applied in the second step of the study to identify, through the analysis of the cause and effect diagram, the risk factors, which will serve as the basis for the response plan to the critical risks identified in the first part of the research. The proposed method intends to improve risk management in software development projects while presenting new research opportunities with the association of two methods that have already been shown to be effective, applied separately, in research from other areas of knowledge.

2 Conceptual Analysis 2.1 Project Risk Management Several organizations use projects not only to create unique products, services, or results through a temporary venture, but also to drive change by reaching specific objectives, thus enabling the transition from a current organizational state to a state future organization (Project Management Institute 2017). However, organizations of all types and sizes need to face the influences of internal and external factors that make achieving these objectives uncertain (ISO 2018). For a project to achieve its purpose, and therefore be successful, undesirable situations such as missed deadlines, budget overruns, poor quality of deliveries, and dissatisfaction of stakeholders must be avoided through effective risk management, with the identification, analysis, and mitigation of the crucial risks to the project (Carbone and Tippett 2004). In this regard, project risk management is treated by PMI (Project Management Institute, 2017) as one of the nine essential disciplines that make up the entire management of a project’s life cycle and has processes for planning, identification, risk analysis, responses and monitoring, can increase the likelihood of positive impacts and reduce the likelihood of negative impacts. PMI consolidates an extensive list of techniques for risk identification (PMI Practice Standard for Project Risk Management 2009) presenting the strengths and weaknesses of each one. Among so many, there are two that are part of the objective of this study, with relevant results concerning risk management in projects: RFMEA and Systems Dynamics, which they will be detailed below. 2.2 RFMEA (Risk Failure Mode and Effect Analysis) The Risk Failure Mode and Effect Analysis in Projects (RFMEA), the modified method from Failure Mode and Effect Analysis (FMEA), is used to identify, quantify and eliminate (or reduce) risks in the project environment. This is defined as a ‘detection technique’

24

L. S. Leite et al.

for risk events with enough time buffer for contingency planning (Carbone and Tippett Carbone and Tippett 2004). Pritchard (2000) first identified the FMEA technique as an advanced format capable of capturing project risks. Carbone and Tippett (2004) extended the technique for planning the prioritization of risks and called it RFMEA (or Risk FMEA). The standard FMEA method evaluates the failure mode of products and services contingent on three factors based on failure; occurrence, severity, and detection, while the RFMEA method considers the same three factors, however focusing on Risks: Probability of occurrence of Risk (how many times it occurred and how valid it is for the project), Impact of the risk (when it happens, what comes below, and how much it affects the project), and Risk Detection (which is the advance in which the risk is detected). 2.3 System Dynamics (DS) J. W. Forrester created systems Dynamics (SD) in 1961. Forrester developed a theory to simulate complex, nonlinear systems that contained multiple feedback loops. He first used this theory to model and analyze industrial problems, such as inventory fluctuations, labor force instability, and falling market share. However, since then, it has been used extensively to several problems from ecological to socio-economic systems. (Marujo 2006). According to Lyneis and Ford (2007), the basic idea behind the method is that any complex situation can be described in terms of elements and flows. Flows are the relationships between the elements. The method’s main focus is the structure composed of the interactions of the elements (flows and levels). This description constitutes the dynamic behavior of the system. Essentially, Systems Dynamics aims to predict a system’s behavior, and to do that, it depends a lot on the use of a model that must contain the complexities of a structure and the multiple feedback cycles to link each element within that structure. The System Dynamics process follows three steps summarized below, and in which only steps 1 and 2 will be adopted in the present study. The construction of different scenarios will be developed in future research. 1. Definition of the problem situation: The problem must be identified; 2. Conceptualization and construction of the model: Define the level of resolution, understanding the influences on the variables and/or elements; 3. Execution of the simulation model and the use of the results: Construction of different scenarios must be analyzed and used to test different situations/decisions.

3 Research Methodology The present study is classified as applied research as to its nature. Exploratory as to the general objective. Case Study and Qualitative as to how to approach the problem, and bibliographic as to the technical procedures. To this end, it will present a proposal for adopting a Critical Risk Analysis and Modeling method in Software Development

Analyzing and Modeling Critical Risks in Software Development Projects

25

Projects, using a combination of two methods that are widely used in the Risk Management area: RFMEA, Risk Failure Mode and Effect Analysis in Projects, for the identification and prioritization of Risks; associated with Systems Dynamics. The Systems Dynamics will be applied in the second step of the study to identify risk factors through analysis of the cause and effect diagram for later elaborate a list of the main recommended actions to address the identified and prioritized risks in the first stage. Therefore, the following questions must be answered during the study: 1. What are the main risks related to software development projects? 2. Among these risks, which are considered critical, identified by the experts of the selected project, which will serve as a basis for applying system dynamics? 3. Among the identified risks, how does the cause and effect relationship between them occur? 4. Finally, what are the risk strategies and recommended actions for the risks identified and modeled in applying the method? A quantitative, descriptive study was carried out through a Survey applied to managers and experts working on the “ANAC Data Search” project of the Brazilian Civil Aviation Agency (ANAC) to meet the research purposes. The survey took place in October 2020, between the 19th and the 28th. The items were structured and adapted from the study by Carbone and Tippett (2004), listed in Sect. 5. Application of the Method. The study also counted on the Vensim simulation software (https://vensim.com/) for the conception of the cause and effect diagram. The proposed method provided the necessary basis for the risk strategies definition and elaborated a list of the main recommendations to address the identified risks.

4 Literature Review No studies related to risk management and software development projects were found using the RFMEA method associated with System Dynamics. Previous studies involving risk management in software development projects have focused on risk identification techniques such as Delphi, complex scenario analysis techniques such as Factor Analysis, literature revision, analogies with economic concepts, proposals for adopting frameworks, and lessons learned from experienced project managers, among others. Only the article that served as the basis for this research, by Carbone and Tippet (2004), uses RFMEA to detect projects’ risks. And the article by Lopes et al. (2015) presented a model of Systems Dynamics involving risk factors that strongly influence software development. Therefore, the application of both associated techniques aims to contribute to future research related to risk management in software development projects and risk management in projects in general. Firstly, the keywords Risk Management, Project Management, and Software Development were used in three different research bases: Web Of Science, Science Direct, and Scopus, with the temporal delimitation of the last ten years, from 2010 to 2020. Secondly, all repeated results were excluded, and 58 items were left, with a sum of 129 citations and h-index 6. Subsequently, this information was applied in VOSviewer; software used to construct and visualize bibliometric networks. The results are presented

26

L. S. Leite et al.

below, where it is possible to verify the terms with the highest number of occurrences found in the selected articles (Fig. 1):

Fig. 1. Bibliometric network of research-related terms

As the intention is to focus on risk management of software development projects in general, all articles that focus on specific software development environments were excluded, such as “Agile Software Development” or “Global Software Development,” leaving 26 articles to be analyzed of which the 13 most relevant articles were listed, based on the number of citations, related to risk management in software development projects: Table 1. Literature review Article, Authors, Year of Publication

Purpose

A contingency estimation model for software projects. Uzzafer, Masood. 2013

Proposed model that considers the estimated cost and risk of software projects to estimate contingency resources

Tool to assess the maturity level of the risk management of a software development process. Gaffo, F. H.; Brigano, G. U. Aoki Horita, F. E. de Barros, R. M. 2013

To present the diagnostic assessment tool that is based on the GAIA Risk framework

Project Risk Management Using the Project Risk FMEA. Thomas A. Carbone e Donald D. Tippett. 2014

RFMEA Model Proposal. Proposal for extending the Failure Mode and Effects Analysis (FMEA) to quantify and analyze the project risks (continued)

Analyzing and Modeling Critical Risks in Software Development Projects

27

Table 1. (continued) Article, Authors, Year of Publication

Purpose

Top Twenty Risks in Software Projects: A Content Analysis and Delphi Study. Sonchan, Pontakorn; Ramingwong, Sakgasit. 2014

State of the Art Review. To present and classify the main risks of the thirty most cited and recently published literature on software project risks

A Probabilistic Software Risk Assessment Proposal for a probabilistic model of software and Estimation Model for Software Projects. risk estimation using Bayesian Belief Network Kumar, Chandan; Yadav, Dilip Kumar. 2015 (BBN) that focuses on the main software risk indicators for risk assessment in software development projects Systems dynamics model for decision support in risk assessment in software projects. Lopes, J. Silva; Braga, J. L. Resende Filho, M. A. 2015

To present a system dynamics model, which is a descriptive technique used for modeling and simulation of systems, involving risk factors that strongly influence software development

A Critical Analysis of Project Management Models and its Potential Risks. Iamandi, O. Popescu, S. Dragomir, M. Morariu, C. 2015

State of the Art Review. Proposal for a bivalent analysis of the current state of risk management within the discipline of software development project management

Proposal of Risk Management Metrics for Multiple Project Software Development. Wanderley, M.; Menezes, J, Jr.; Gusmao, C. Lima, F. 2015

Proposal to use a strategy based on metrics as a tool to support multiple project managers with an emphasis on risk factors

Research on Risk Analysis and Management To present the methods of processing software in the Software Development Process. risks and the means of estimating and managing Huang, Quanzhou. 2015 risks, analyzing mainly the relationship between software risks and models of development processes A Fuzzy method for qualitative analysis of Classify the risks in a software development risks in software development projects. Lima project and measure the project’s exposure to de Campos, L. M.; Lima, A. S. 2016 risks, using two metrics based on fuzzy values: risk exposure index and risk scores Preeminent Risk Factor Affecting Software Development. Gondal, H. A. H. Fayyaz, S. Din, S. M. U. Zeb, M. D. Nadeem, B. 2018

Provide a comprehensive analysis of the risk factors that may occur during each phase of the software development project life cycle

Risk factors in software development State of the Art Review. identify and map risk projects: a systematic literature review. factors in software development project Menezes, J, Jr.; Gusmao, C. Moura, H. 2019 environments A Strategy Using Continuous Simulation to Present a simulation as a strategy to mitigate Mitigate Effort Estimation Risks in Software failures in the estimation of the software Projects. Zanella, J.; Montenegro, J.; Canto, development effort J.; Victoria, J.; da Costa, C. 2019

28

L. S. Leite et al.

After analyzing the 13 articles listed in Table 1, those articles that focused on identifying the main risks of software development projects were selected: Ramingwong and Soncha (2014); Lopes et al. (2015); Gondal et al. (2018) and Menezes Jr. et al. (2019) for the construction of Table 2: Table 2. Main risks related to software project management

The risks to be considered will be present in at least three of the selected texts. They were used in the application of the Survey with software development project experts.

5 Method Application: Combination of RFMEA and System Dynamics Considering the 23 risks in Table 2, ten were selected for the Survey with the experts of a large project at the Brazilian Civil Aviation Agency (ANAC), a regulatory agency that aims to standardize and supervise civil aviation activity in Brazil, both in its economic aspects and about the technical security of the sector. • Risk 1. Unrealistic schedules and budgets: Inefficient planning in project deadlines and budgets that directly impact the project milestones. • Risk 2. Continuous requirement changes: constant changes in functional and nonfunctional requirements or inclusion of unplanned features that directly impact project deadlines and budget.

Analyzing and Modeling Critical Risks in Software Development Projects

29

• Risk 3. Developing the wrong software functions: features that do not meet user expectations due to failure in some design stages. • Risk 4. Gold Plating: the addition of any feature not considered in the original scope plan or product description. • Risk 5. Developing the wrong user interface: The user interface does not meet user expectations due to failure in some design stages. • Risk 6. Inadequate knowledge/skills and other deficiences: Risk related to the technical incompatibility of those involved, required for the project’s development. • Risk 7. Resource usage and performance: Deficiencies found in the acquisition and/or use of information technology resources such as hardware, software, networks, databases, among others. • Risk 8. Shortfalls in externally furnished components: part of the software solution provided by a third party with unexpected performance • Risk 9. Shortfalls in externally performed tasks: project activities attributed to external teams or partners carried out incorrectly or incompletely. • Risk 10. Real-time performance shortfalls: Instabilities in the system performance, for example, the occurrence of delays in the system response time. 5.1 First Step RFMEA: Identifying Project Risks The project selected to qualify the risk factors aims to offer the Brazilian Civil Aviation Agency (ANAC) a solution that allows, in a centralized, wide, safe, and reliable way, data analysis and mining, integrating and listing the approximately 120 information systems implemented, together with an invaluable amount of unstructured data, such as spreadsheets and documents, sensitive to the organization’s decision-making process. The project’s objective is to promote easy and democratized access to this set of data, which will be organized in a Data Lake, a repository of structured data and unstructured data, with predefined views and that favor the search through keywords. The application will be built based on Elastic Search technology. Since RFMEA requires the risk’s value materializes, it is important to highlight that the project term spanned August to October/2020, totalizing 63 working days. Besides this, the project’s direct cost was estimated at R$ 21.240.00 (twenty-one thousand two hundred and forty reais). The questionnaire was structured according to the Project Risk FMEA method proposed by Carbone and Tippet (2004). The monetary values and the estimate of three other variables for each Risk were included: 1. The probability of occurrence (Fig. 2); 2. The impact if it occurs (Fig. 3) and 3. Detection of the risk or how difficult it is to detect it (Fig. 3). Each variable used a scale of 5 ranges, as described below:

Fig. 2. Likelihood value guidelines Carbone and Tippett (2004)

30

L. S. Leite et al.

Fig. 3. Impact value guidelines e detection value guidelines Carbone and Tippett (2004)

The Survey was carried out in Google Forms between October 20 and 28, 2020. It was applied to business experts and project technicians, composed of software system managers, product owners, technical representatives, managers, architects, and project facilitators. The purpose of applying the questionnaire was to collect information related to Step 2 of the RFMEA Method. The information collected was used to calculate the arithmetic means of each of the 30 questions to have a single value each. Then, still in compliance with Step 2 of the RFMEA, the individual Scores (probability * impact) and RPN (probability * impact * detection) of the ten estimated risks were calculated. With this, the following scenario was obtained, after the ordering of risks through the values of RPN and Score (Table 3 and Fig. 4): Table 3. Risks of ANAC Data Search project ordered by RPN and Score ID

Risks

RPN

Score

1

Unrealistic schedules and budgets

414

72

6

Inadequate knowledge/skills and other deficiences

314,02

43,31

7

Resource usage and performance

259,88

57,75

10

Real-time performance shortfalls

252,34

53,13

8

Shortfalls in externally furnished components

249,38

33,25

9

Shortfalls in externally performed tasks

191,40

34,8

3

Developing the wrong software functions

210

35

2

Continuous requirement changes

171

38

5

Developing the wrong user interface

87,75

14,63

4

Gold plating

84,91

17,88

Analyzing and Modeling Critical Risks in Software Development Projects

31

Since the RFMEA methodology also includes steps 4 and 5, the Pareto Graphs of the RPN and Score risk values were shown below (Fig. 4):

Fig. 4. Pareto graphs of risks

After that, it was elaborated the Risk Dispersion Graph referring to RFMEA Step 5. Then, it was possible to identify the risks with the highest Score and RPN values by observing the first quadrant, thus classified as critical project’s risks, essential for the study of its causal relationship with Systems’ support Dynamics modeling.

Fig. 5. Project risk dispersion graph

In Step 6, RFMEA guides the analysis of the risks listed in Fig. 5. The risks of higher scores values are analyzed concerning the risks of higher RPN values. If they occur, even in the initial stages, it’s possible to decide on the project’s discontinuity. In this research, Step 6 only was used to ratify the main risks found and shown in Fig. 5. Their relationships were analyzed through the construction of a System Dynamics causal and effect diagram. Besides that, the last RFMEA steps (7, 8, and 9) were not the target of this research. 5.2 Second Step: System Dynamics For applying the second step of the research, Vensim, a visual modeling tool, was used. This tool allows development, document, simulation, and analysis models of causal relationships of Systems Dynamics. It is possible to observe in Fig. 6 the cause and

32

L. S. Leite et al.

effect diagram of the risks with the highest RPN and Score: Inadequate Knowledge and Skills and other deficiencies, Unrealistic Schedules, and Budgets; Resource usage and performance and Real-Time Performance Shortfalls.

Fig. 6. Cause and effect diagram with critical risks

During the modeling of the diagram, it was necessary to identify the causal variable, that is, in the modeling process, it was identified that Risk 6 “Inadequate Knowledge and Skills and other deficiencies” directly interferes in the Project Schedule and Budget, making them unrealistic (Risk 1). The time and budget’s increase impact the execution of the project, and “Resource usage and performance” (Risk 7) are more recurrent due to the cost of acquisition and maintenance of these IT assets, which therefore increases the probability of occurring “Real-Time Performance Shortfalls” (Risk 10). Thereafter, the diagram was evolved with the inclusion of cause and effect relationships of the risks related in Fig. 7 with the ones selected in the literature review.

Fig. 7. Cause and effect diagram with the 10 main risks

Analyzing and Modeling Critical Risks in Software Development Projects

33

Based on the model initially proposed, it was verified that the risks “Shortfalls in Externally Performed Tasks,” “Developing the wrong User Interface,” “Developing the wrong Functions and Properties,” and “Gold plating” are related to the critical risk “Inadequate Knowledge and Skills and other deficiencies” since the more internal and external human resources to the project team with inadequate knowledge and skills for the execution of the project, the greater the probability of errors in the project. On the other hand, the risk “Continuous Requirement Changes” is directly related to “Resource usage and performance” and “Unrealistic Schedules and Budgets,” as constant changes in functional and non-functional requirements or inclusion of features not initially planned can impact project deadlines and costs, as well as generating the need for readjustment of computational resources. Finally, the risk relationship “Resource usage and performance” with “Real-Time Performance Shortfalls” already explained in the first model (Fig. 6) received the risk connection “Shortfalls in externally furnished components” demonstrating that problems related to malfunction of solutions provided by other external project teams directly interfere with the performance, costs and project schedule, in addition to impacting the need for the use, acquisition, and maintenance of internal resources. Figure 7 shows a positive link between the cause and the effect, which means that if the cause increases, the effect also increases and remains to be planned. On the contrary, if the cause decreases, the effect in the same way. An increase in personnel with inadequate knowledge and skills indicates a possible increase in the chances of the project schedule and budget not being fulfilled, and so on. The bottom line is when the cause is a flow risk that accumulates. It is also true that the effect is attributed to other risks.

6 Conclusions In order to obtain a greater probability of success in software development projects, regardless of the context in which it is inserted, this study proposed to apply a combination of two methods: RFMEA and System Dynamics, to analyze the main risks related to software development projects, searched in the literature review, and after that, to model the cause and effect relationship among them. The priority of each risk was determined using two criteria. The first criterion used was the ‘Dynamic Ripple Effect’, which is the effect that a given risk causes, when triggered, on another risk that has a direct relationship. This criterion is collected in the analysis of the System Dynamics of the project risks and, according to the multiplicity and relational severity of the risk, it is classified in five ranges, as follows (Table 4):

34

L. S. Leite et al. Table 4. Dyanmic Ripple Effect (DRE)

In cases where a risk covers more than one range, its DRE is defined through the highest criticality. The Dynamic Ripple Effect was defined as the first prioritization criterion due to the effect that occurs in the event of a risk, occurring in possible cascade occurrences in the other risks and also to circumvent likely imprecise probability and impact (RPN) of the risk by the project team. The second criterion is the relationship between probability, impact, and risk detection called ‘RFMEA RPN’. After obtaining the two criteria, the Preliminary Risk Response Plan (Table 5) with the required risks is then elaborated, which is the result of the Case Study, allowing then to identify the risk factors. Table 5. Preliminary risk response plan

Analyzing and Modeling Critical Risks in Software Development Projects

35

This risk response plan aims to indicate the risk factors related to each of the main risks of software development projects and recommend actions for each of them. The recommendations should serve as the basis for a more detailed risk treatment plan. It is important to remember the risk management must be carried out from the beginning to the end of the project. Combining two methods for managing and analyzing risks, such as RFMEA and System Dynamics, provides a more detailed evaluation to understand which risks the project manager should be concerned with. It is recommended that future studies be carried out to quantitatively analyze the proposed method, involving a more significant number of projects in different contexts, scenarios, degrees of uncertainty, and complexities, with a more in-depth approach of System Dynamics, so that the list of the main risks related to software development projects can be revalidated, using in addition to System Dynamics, other methods such as Real Options Theory.

References International Organization for Standardization. ISO. ISO 31000:2018: Risk management — Guidelines (2018) Carbone, T.A., Tippett, D.D.: Project risk management using the project risk FMEA. Eng. Manage. J. 16(4), 28–35 (2004) Project Management Institute, 2009, PMI Practice Standard for Project Risk Management (2009) Project Management Institute, 2017, PMBOK: Project Management Body of Knowledge (2017) Pritchard, C.L.: “Advanced Risk — How Big is Your Crystal Ball?” Proceedings Project Management Institute Annual Seminars & Symposium, September 2000 Marujo, L.: Um Modelo Estimativo do Retrabalho utilizando Dinâmica de Sistemas para Gerenciamento de Projetos (2006). https://www.researchgate.net/publication/255651297 Lyneis, J.M., Ford, D.N.: System dynamics applied to project management: a survey, assessment, and directions for future research. Syst. Dyn. Rev. J. Syst. Dyn. Soc. 23(2–3), 157–189 (2007) Lopes, J., Braga, J., Resende, F.M.: Systems dynamics model for decision support in risk assessment in software projects. J. Softw. Evol. Process. 27. n/a-n/a. https://doi.org/10.1002/smr. 1754 Ramingwong, S., Sonchan, P.: Top Twenty Risks in Software Projects: A Content Analysis and Delphi Study (2014). https://doi.org/10.1109/ecticon.2014.6839820 Hafiz Ali Hamza, G., Saira, D., Sajida, F., Muhammad, Z., Beenish, N.: Preeminent risk factor affecting software development, pp. 1–7 (2018). https://doi.org/10.1109/icacs.2018.8333492 Júlio, M.J., Cristine, G., Hermano, M.: Risk factors in software development projects: a systematic literature review. Software Qual. J. 27, 536 (2018). https://doi.org/10.1007/s11219-018-9427-5

The Communication Between Client-Developer in the Process of Requirements Elicitation for a Software Project Sebastián Alvarez1(B) , Kevin Duy1 , Mireya Zapata2 , Jorge Galarza1 , Danilo Martinez1 , and Carlos Puco1 1 Universidad de las Fuerzas Armadas ESPE, Avenida General Rumiñahui S/N, Sangolqui

171103, Ecuador {jsalvarez4,kjduy,jlgalarza3,mdmartinez,capuco}@espe.edu.ec 2 Research Center of Mechatronics and Interactive Systems - MIST, Universidad Tecnológica Indoamérica, Machala y Sabanilla, Quito, Ecuador [email protected]

Abstract. Communication is a fundamental part of the requirements elicitation process for a software project, but the constant changes in the requirements make it difficult to complete a project successfully, and sometimes the customer’s expectations are not met. The problem of communication has different reasons, among them, the shortcomings in the process by which the requirements engineer contacts the customer and the customer does not always have a clear idea of what he really needs. In order to identify the reasons for the communication problems between the requirements engineer and the customer, we have carried out a survey with the participation of employees from two software development companies. The results show the comparison and contrast of the effectiveness of the traditional processes against the current processes in the same way, we can highlight the preference of the experts when choosing a method of elicitation against another one and the advantages and disadvantages that each one represents. We also get results about how to improve the processes that present the greatest number of problems. In addition, it is worth mentioning that results were also obtained regarding how professionals recognize when there is a problem between the developer and the client or vice versa and the possible solution that should be applied to avoid some kind of conflict in the medium or long term in the software’s development project. Keywords: Software engineering · Requirements elicitation · Communication in software development · Software projects

1 Introduction An important stage in Software Engineering is the requirements elicitation (sometimes referred to as “requirement gathering”) for a system since it marks the starting point for other activities of the project and allows to verify if the proposed objectives are achievable with an optimal level of quality to get the user’s satisfaction. The requirements elicitation © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 36–45, 2021. https://doi.org/10.1007/978-3-030-72654-6_4

The Communication Between Client-Developer

37

refers to the identification and documentation of these requirements in a system, in other words, “what the system has to do” according to what a certain client or project Stakeholders propose. In this stage of development several problems can be presented, Davey et al. [1] affirm that there are 9 main problems in the requirements elicitation one of them is the communication between client and developer which is the main topic of our research. The authors provided suggestions to avoid difficulties in the process and to keep the project on track such as the use of procedural directions when asking questions. These encourage further reflection before responses and have been tested several times with measurable success. The research arises from the desire to highlight this shortcoming and how it crucially affects the project to be carried out [2] and the people who are part of it and the resources invested in it giving more prominence to this issue when carrying out such work. Experts on the topic [3, 4, 5] say that this is one stage of software projects in which more mistakes are made mainly in communication which can end in the project’s failure because of a bad definition, specification and management of requirements. For this reason the main aim is to produce an paper that deals with the communication problems of this stage which will be a contribution to future requirements engineers that try to solve problems in the requirements elicitation where they know techniques to understand what the client wants and to capture the proposed requirements. The paper is structured as follows: Sect. 2 deals with related work. In Sect. 3 we present the method that was used. We show the results in Sect. 4 and the discussion of these in Sect. 5. Section 6 shows the threats of validity of our study. Finally, Sect. 7 shows the conclusions.

2 Related Work Abelein et al. [6] conducted a series of semi-structured interviews with twelve experts working in IT business coordination and described their experiences gained from 69 large-scale IT projects. The analysis of their results showed that communication with the developer is limited and that there is no commonly used method for user-developer communication in the design and implementation activity of IT projects. This research is like what we intend because its results go to contrast with ours despite not using the same method of data collection. Alvertis et al. [3] present a development method called “Cloud Teams” which shows that the physical presence of the end user in any part of the development process of a system (including the process of requirements gathering) is not essential to do a good job. Many of the deficiencies related to the requirements elicitation can be solved in this way and thus avoid certain causes where we intuitively find the problems at the time of collecting requirements. To the being a different and innovating perspective, it provides valuable information and results to us for our research, getting that the same one can be directed towards possible correct results. Lopez et al. [4] provide us relevant information about the traditional processes of requirements gathering which do not consider communication problems. The authors address the traditional processes and their bad performance, their conclusions show

38

S. Alvarez et al.

that the difference of languages and time zones is a key point in the failures of the communication between customer-developer because the links of understanding the requirements are cut. We can also observe that the data collection method used for this research is the survey. Romero et al. [5] present information about the reasons why failures happen during the requirements gathering process. The aim of the study is to present certain solutions to equalize the failures that this process has providing its results to continue and improve them. The authors present proposals such as the development of a simulator for students where they are trained in obtaining requirements before going out to the work area. The simulation of the behavior of the interested parties contemplates cultural, cognitive and geographical aspects as well as the time differences. These researches are very important to us because of their scientific relevance and their closeness to our research. Their results and contributions feed our work giving us the possibility of not starting “from scratch” but to have certain advances from which to continue or at least a marked beginning.

3 Methodology We applied the case study guidelines in software engineering by Runeson et al. [7]. The aim of our study is to propose a solution/improvement in the communication between client and developer through the application of methodologies that allow the intermediary (requirements engineer) to fully understand the client’s needs and express them correctly to the development group. The context considered is the software engineering industry, we have 2 units of analysis as shown in Fig. 1.

Fig. 1. Cases and units of analysis [7]

Currently, the two units of analysis we have chosen have some problems at the time of researching the requirements and this is not due to lack of experience of the

The Communication Between Client-Developer

39

requirements engineers but is linked to other factors including communication between client and developer and this problem may be directly related to the method used to collect data. We want to find out how these problems can be avoided and give way to an optimal way to obtain the requirements that the client wants. For our research we have 2 unit of analysis in the same context: Unit of analysis 1: UA1 is an agile software consulting firm where the team works for a common goal to add value to the clients’ business in each project, they perform through high-quality software solutions. They expect clients to have a unique experience by providing them with high-quality code and exceptional service. The vision of the consulting firm keeps them on the path of constant professional growth, and this is reflected daily in the company’s culture. The portfolio includes projects such as a ticketing platform for the tourism industry, the complete construction of a mobile banking system, and the improvement of financial platforms with high user traffic and other interesting projects. Unit of analysis 2: UA2 is a marketing agency to help marketing and technology leaders connect the dots in the customer journey. They design and create web, mobile, and business experiences that bring brands and their customers closer together. 3.1 Research Questions The research question of our study is: RQ: How to improve the communication action between client - developer in the process of requirements elicitation? The research question is very broad; therefore, it is necessary to break it down in order to obtain more specific and objective results, so we have the following research sub questions: 1. How important is the process of requirements elicitation for the development of a software project? 2. Are the traditional processes of requirements elicitation optimal? 3. Which are the most used techniques by professionals nowadays to get requirements? 4. Can a requirement gathering technique fail? 5. How could the process or processes that present failures be improved? 6. How do you identify when there is a problem between the requirements engineer and the customer? 7. How can problems between the developer and the customer be solved? 3.2 Data Collect According to Linares-Fontela [8] the first step is to decide if a survey is the right tool for each situation. To answer our research question, one of the most viable tools is the survey. The survey comprises five closed questions and two open questions so that the answers got can be analyzed and represented in graphs.

40

S. Alvarez et al.

Interviewers must decide exactly what type of information they want to find in order to decide what type of survey is appropriate for each situation. Based on our research problem, we determined that the best survey to apply is a detailed survey [8] since it will help us collect accurate data and be able to interpret and analyze it through graphics and statistical calculations. Linares [8] proposes three steps to design an effective questionnaire or survey. 1. Define the scope of the information to be got. 2. Determine the order of the questions 3. Decide what kind of question forms will get the necessary information. The software that we used was the Google Forms tool to carry out the survey with which the necessary information for the seven questions was collected. It is worth mentioning that the two open questions we had should have been analyzed.

4 Results In this section we present the results got through the survey applied to workers in the software industry who are collaborating in this study. 4.1 The Importance of Requirements Elicitation for the Development of a Software Project The requirements elicitation is important because the requirements engineer needs to elaborate the documents or diagrams that are desired to make about the new project since with the elicitation the criteria can be extracted as the requirements of vital importance, the idea that the client has about the expected final product and mainly the real requirements can be identified to be able to anticipate and to control the changes of requirements during the development of the project [5]. Based on the above, we can affirm the importance of the requirements elicitation, since they are the foundations that will support the development of the software project to direct it towards the customer’s expectations. The results achieved confirm that it is of utmost importance getting requirements to develop a software project; it is a process which sets the basis for any process to be carried out. 4.2 The Traditional Processes of Obtaining Requirements Are not Optimal In order to know if the processes we apply now of gathering the requirements are effective, the question is posed in the survey. The 77.8% of developers who answered the survey show that getting requirements is not optimal, compared to 22.2% who say it is, showing that most of the respondents feel the central problem being that it should change or innovate in some factors.

The Communication Between Client-Developer

41

4.3 The Most Used Techniques Nowadays to Gather the Requirements To develop skills in the requirements elicitation in software engineers and students it is proposed to use simulation techniques and artificial agents [5]. But no matter what kind of technique is going to be used, the previous or later communication with the customer will be always important and this area must be reinforced in the requirements engineers. One of the most used technique to carry out the requirements gathering is Focus groups, this can be directly related to including agile methodologies where the entire work team is immersed with the client to carry out the requirements gathering. The next is the direct observation, however, unlike the focus group, a professional does not know or apply this technique, we can assume that this may be by instruction of your project manager or because it is not applicable to the project, they are performing. Then there are the surveys, which have fewer professionals who know or apply this technique, the main reason for their lack of knowledge may be because they do not apply it within their work environment or because it is very expensive and takes a long time to perform. Finally, we have the experiments which are the ones that fewer people know about, and like the surveys, may be due to the high time invested or because the costs of conducting them are very high. 4.4 The Failures of a Requirements Gathering Technique As we know that no process is free of failures, in our study, we identified that the survey is the one that stands out the most. This technique is directly related to the lack of communication between the requirements engineer and the client, because the surveys may contain closed questions, which considerably diminishes the clients’ criteria or because there may not be questions in which the client can give his opinion about the requirements he had. Then, there are the experiments, which, being a relatively new technique in software engineering, are the least reliable for clients and professionals, which can lead to errors within the requirements education process. It is followed by the focus groups and because in the meetings that are held between all team members, there may be factors such as shyness or lack of participation within the meeting, which makes the process of requirements gathering difficult or even null because there is not good communication. Finally, the one that generates fewer failures is the direct observation, since here the requirements engineer is a direct witness of what is necessary (the requirements) to develop the project. 4.5 Improvement of Faulty Processes To improve the processes that present failures, it is necessary to make efforts in the globalization that currently governs society, in the software project teams and to know how to adapt according to the geographical location that we are in the world [4]. Globalization is referred to with the objective that we must grow with the world and not get stuck in traditional methods but use newer and more effective ones, although it tells us that everything will depend on the geographical location in which we are and this to know how to apply traditional or current methods according to the country we are in and it is believed to be more appropriate to be used.

42

S. Alvarez et al.

In our survey we got different opinions on how to improve this lack of communication with the client, so several respondents concluded that it is not enough to carry out the traditional processes although it is done correctly, it is necessary to have a deeper analysis with the client taking the information in a concrete and conscientious way. 4.6 Identification of a Problem Between the Requirements Engineer and the Customer As an answer to how to detect if there is a communication problem between the client and the developer, we have that 77.8% of the developers to participated in our study claim to have communication problems when the client presents constant changes in the system requirements. The 22.2% have had experience with the little interest the client usually shows in the development process. While there are no respondents who claim that the client has shown discomfort. We can conclude that there are communication problems with the client in the process when there is a constant change of requirements and clients do not show discomfort in the process although there are communication problems with the client. 4.7 The Viable Solution of the Problems Between the Developer and the Client The feasible solution to this problem of communication according with 42% of the surveyed people is that the developer should intervene by suggesting various ideas for the project to be carried out based on experiences and judgments that have given excellent results in previous works. Although this question may depend on several factors, among them the cause that is generating the problem is of vital importance to detect it and apply the respective solution. However, a process proposal can be put forward that is based on an offshore distribution and division of roles, that is, outsourcing [4]. This solution is generic and although it is not ideal for all cases, it can be applied to greatly reduce the existing problem, because hiring a staff outside the company and being totally impartial to the requirements engineer and the client can give an opinion that is causing the problem and its respective solution, although this will generate more costs to the company, can be a viable solution to correct the miscommunication between client - developer. Finally, to answer the research question it is necessary to highlight the existence of multiple factors that determine the success or failure of the communication between client - developer during the initial phase of a software project, because it is necessary to take into account the strategic, cultural, adequate communication, management knowledge, project management processes and technical problems that may arise [5]. For this reason, to improve communication it is necessary to first focus on the strategy to be used, that is, which method of requirements elicitation will be used. Keep an open mind to avoid cultural problems with the client. Always maintain an environment of respect and good communication with the customer. One aspect to highlight is that the requirements engineer must be a good communicator. Remember the requirements engineer must not only dedicate himself to the requirement elicitation but also, to know other aspects of project management such as its duration, cost, among others. And last but not least, he must know how to use a technical language if the client is able to understand what the

The Communication Between Client-Developer

43

developer is talking about. If this is not the case, he must maintain the same capacity to express the technicalities of the project in a natural language.

5 Discussion In this section, we present a discussion of the main findings of the study. The 100% of the surveyed stated that the process of requirements gathering in software development is one of the most important phases of the project. As it is a high percentage, all the surveyed professionals have carried out this process and at some point, they have presented problems that lead to the failure of the project or to the dissatisfaction of the client with the developed system. Therefore, it is essential to carry out this process correctly before developing the system. A high percentage of participants (77.8%) no longer consider the traditional processes of requirements gathering such as questionnaires to be optimal. This factor may be because of the technological advance that developers face every day, since the functionalities required by clients are increasing, making the system more complex. As the system is more complex, the requirement gathering will be more extensive and must be done with greater thoroughness, using new techniques and improving communication with the client. The “Focus Group” technique is preferred by a significant group of professionals surveyed because of its agility. However, Barrón et al. [9] specify that sometimes it is necessary to use over one technique. It is also mentioned that the different requirement gathering techniques that can be used often vary depending on the needs of the client and the complexity of the system. Customers show signs of problems in the requirements gathering process for various reasons, because of direct customer dissatisfaction or various indirect factors such as constantly changing requirements in the system. The probable reason that the customer wants to change functionalities from one moment to another is the lack of trust with the developer, so it is advisable to repeat the requirements gathering process from time to time, verifying that the required functionalities are the ones implemented. Identifying this communication and trust problems with the client and correcting them in a timely manner will save development time and time in error correction or constant changes of functionalities. A challenge for the developer is to have the client as the aim and fundamental piece in the system’s development, because only the client knows what he wants for his product and its functionalities. For this reason, communication must generate security, trust and in a certain way create a link that allows communication to flow agile, making the client’s needs known in a clearer way. With this, the developer will be able to implement the requirements in an optimal and efficient way. The results arrived at in this research may be comparable to the results got in another research, such as Abelein et al. [6] conclude after their research that the communication between client - developer does not exist in the long term within the process of development of IT projects, coinciding with our research. The data collection method used by Abelein differs from ours, since for his research he conducted personal interviews, while in ours a general survey was conducted. This can reinforce our conclusion because from different data collection methods we can come to the same conclusion.

44

S. Alvarez et al.

Hassan et al. [10] provide us with a different perspective on customer-developer communication, because it deals with how people (customers) interact regarding improving or increasing a free application in the mobile application store and comes to the general conclusion that customers do not provide the information needed to make improvements to the applications. This can be expressed as a communication failure between the users of the applications and the development team of the same, so a correct feedback is not given, and the developers must find the failures or improvement needs by themselves. This shows us that the result of our research is like the present one, since, as we mentioned before, there is not a correct communication between client and developer.

6 Threats of Validity The results got in this research have some threats of validity due to different aspects found throughout the research. In a certain way, the results got may have a certain bias because when the research was started; the results obtained were considered as possible. In addition, the following two threats should be considered: The lack of experience of the adjunct researchers: the present research is affected by the lack of experience of the adjunct researchers since this is the first time they are conducting a complete research. The small number of professionals surveyed- performing research in times of pandemic makes it difficult to have professionals available for research, so the results of our research result from the analysis of a small study population.

7 Conclusion The results exposed in this research emphasize that the requirements elicitation is fundamental in a software project almost as important as the program itself. According to the answers got in the surveys carried out, the fundamental problem is in the communication, in the exchange of ideas, since there is a flaw that makes the project does not comply with the estimated time for its realization. A fluent and reliable language must be used between the requirements’ elicitor and the person responsible for the project. This is very important, and the person in charge of the requirements would be able to contribute with ideas that have been successful in previous works. The knowledge and experience are an important point to consider in order to conceptualize in a more solid way the idea of the product. With this work we have clear that there is a flaw in the process of requirements gathering which could derive in other more specific problems or of the same nature which can be studied from different points of view as it has been done in this research.

References 1. Davey, B., Parker, K.: Requirements elicitation problems: a literature analysis. In: Proceedings of the 2015 InSITE Conference, Informing Science Institute, p. 910 (2015) 2. Chen, W.N., Zhang, J.: Ant colony optimization for software project scheduling and staffing with an event-based scheduler. IEEE Trans. Softw. Eng. 39, 1–17 (2013)

The Communication Between Client-Developer

45

3. Alvertis, I., Koussouris, S., Papaspyros, D., Arvanitakis, E., Mouzakitis, S., Franken, S., Kolvenbach, S., Prinz, W.: User involvement in software development processes. Procedia Comput. Sci. 97, 73–83 (2016) 4. Lopes, L., Prikladnicki, R., Audy, J., Majdenbaum, A.: Requirements Specification in Distributed Software Development-A Process Proposal 5. Romero, M., Vizcaíno, A., Piattini, M.: Developing The Skills Needed For Requirement Elicitation In Global Software Development 6. Abelein, U., Paech, B.: State of practice of user-developer communication in large-scale IT projects results of an expert interview series. In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). pp. 95–111. Springer, Cham (2014) 7. Runeson, P., Host, M., Rainer, A., Regnell, B.: Case study research in software engineering: Guidelines and examples. Wiley (2012) 8. Linares-Fontela, J.: Guía para diseñar encuestas. In: El Logro Del Equilibrio En Las Microfinanzas, pp. 53–55 (1996) 9. Barrón Sánchez, F., Montes Venegas, H.A., Romero, M.: Técnicas para el levantamiento de requerimientos en el desarrollo de un sistema de información. Pist. Educ. 36 (2015) 10. Hassan, S., Tantithamthavorn, C., Bezemer, C.P., Hassan, A.E.: Studying the dialogue between users and developers of free apps in the Google Play Store. Empir. Softw. Eng. 23, 1275–1312 (2018)

Unreadable Code in Novice Developers Daniel Avila1(B) , Edison Báez1 , Mireya Zapata2 , David López1 , Diego Zurita1 , and Danilo Martínez1 1 Universidad de las Fuerzas Armadas ESPE, Avenida General Rumiñahui S/N, Sangolquí

171103, Ecuador {dravila2,epbaez1,dalopez18,dazurita3,mdmartinez}@espe.edu.ec 2 Research Center of Mechatronics and Interactive Systems - MIST, Universidad Tecnológica Indoamérica, Machala y Sabanilla, Quito, Ecuador [email protected]

Abstract. Throughout his degree, a developer must face multiple challenges, one of the hardest is to understand unreadable code. This paper trying to identify the cause for the lack of usage of coding standards on students, throughout a survey carried out among first and second levels of degree of the Software Engineering. As well as acknowledging signs that would suggest an individual is prone to develop bad coding habits. Keywords: Code · Unreadable · Lack · Standards · Root · Cause

1 Introduction In the world of software development, the use of programming standards is a highly recommended practice that experts tend to focus a lot on. Most programming standards aim to improve maintenance capability as well as guaranteeing that a shared view of all team members is maintained throughout the code or at least for the most part of it. Besides, writing readable code greatly facilitates its understanding for any foreign developer [1]. However, developers are primarily motivated to develop new applications features, but they are not excited to refactor the source code or implement better test cases [2]. Improving the quality of the code is not deem as something fun to do, but rather, it is regarded as a boring or even tedious task. This lack of motivation for the use of coding standards exhibit a latent problem among developers. Mimicking the partitioners’ behavior, students strive to learn multiple programming languages, to learn to program various software functionalities. Yet, the question arises if students put the same amount of effort into learning programming standards, since bad programming habits are often developed in the early years of academic training. Bad programming habits is often the cause for code smells, and code smells denote symptoms of poor design and implementation choices [3]. Among these bad habits, the most repeated one is writing illegible code, i.e. functional code but only understandable to the person who wrote it. According to Martin [4], the illegible code may work, but © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 46–51, 2021. https://doi.org/10.1007/978-3-030-72654-6_5

Unreadable Code in Novice Developers

47

if there is no clean code the development process can become a nightmare. Every year, countless hours and resources are lost due to misspelled code. If the consequences of writing unreadable code are well documented, then why students continue to write it. To try to clarify this question this research work aims to identify students’ thoughts on the importance of using programming standards and determine whether they are aware of the impact this will have throughout their academic training and eventually on their working lives. For this research, we had the collaboration of students from introductory courses of the Software Engineering Bachelor Program, who by conducting a survey expressed their beliefs regarding aspects such as: • Importance of applying programming standards. • Benefits of applying programming standards. • The problems of writing unreadable code. This paper is structured as follows. Section 1 introduces the subject. Section 2 covers an analysis of related works. Section 3 provides the design and explanation of how this study was conducted. Section 4 contains the results obtained and Sect. 5 the discussion generated by them. Finally, the conclusions of this study are presented on Sect. 6.

2 Related Works A study was conducted to 33 Professors of Object-Oriented Programming at the University of Trás-os-Montes and Alto Duro, who were asked about the importation of good coding practices of which 32 answered affirmatively. It should be noticed that these professors while assessing their students work gave more relevance to the code readability than the code functionality. Indicating that the more readable the code, the easier it was for the professor to evaluated it. In addition, it was concluded that there is a pseudo hierarchy within good practices, where some are considered more “necessary” than others [5]. Good coding practices are an essential component of introductory courses in software engineering and might lead to better coding skills, knowledge of and compliance with coding standards, and peer-to-peer communication within equipment [6]. This practices further helps to raise team awareness and teamwork among students, something essential that should be encouraged on universities throughout the whole career. Moreover, refactoring a teammate’s code creates an inner sense of code maintainability and can be a challenge that delivers great results in the developer skills improvement. According to Neto et al. [7] interest in the study of computer science has been increasing which is the reason that test concepts have been implemented in basic coding classes. In his study he discusses the new POPT programming approach (problem-oriented programming and testing) for introductory programming courses. Through the use of what he called blind tests it was concluded that POPT stimulates the implementation of better quality programs, supported by the fact that the version of a program that followed POPT guidelines approved twice as many test cases (defined by the lecturer) than the non-POPT version, being both versions written by students.

48

D. Avila et al.

Issues such as unreadable and non-reusable code, high defect density, and high maintenance costs are changing the mindset of managers working in software companies. The best place to solve these problems should be an educational environment such as the university, where students begin to write code on their initial classes [8]. Meaning that programming standards must be taught early in the career path, so that it does not bring complications at mid or long term. To allow students to easily understand and utilize coding standards, a simplified version of easy-to-use coding standards must be designed. Compliance with coding standards is becoming an important assessment index of a software engineer’s teamwork capacity. The ability to cooperate in a development team and take part on code readability matters that are also related to code maintenance is a key factor in the overall quality of the software.

3 Study Desing To carry out this research, the case studies technique was used following the guidelines of Runeson et al. [9] that proposes to recreate real-world situations in a laboratory in order to study a specific phenomenon. Case studies are generally applied in research or descriptive and exploratory studies. However, they are not discarded in other types of studies. The purpose of this study is to determine what criteria students have about the use of programming standards. That is, it seeks to establish the level of relevance that students give them within their academic training. Programming standards are conventions that determine how code is written according to the programming language been used. A programming standard will tell us how to declare variables, structures, tables, as well as how objects should be defined. Programming standards are important for many reasons, including the following: • So that the programmer can easily remember the name of a variable. • So that programmers can read and understand quickly, code written by other programmers. • For code to be readable, it is important to use the same conventions throughout the program. It is imperative that an inexperienced programmer recognizes the importance of good practices in the writing of his code so that he can develop good habits that allow him to function properly in his professional life, especially when working within a development team. 3.1 Units of Analysis and Use Cases According to Yin [10] integrated case studies have different analysis units depending on the research question and the very dynamics of the research. Taking this into account, the following integrated case studies will be carried out in this research work:

Unreadable Code in Novice Developers

49

Case1: Methodology of teaching and learning programming standards for novice students of Software Engineering. Case2: Students from the first two levels of the Software Engineering Bachelor Program. 3.2 Research Questions We have posed the following research questions: RQ1: Do professors consider the readability of the code when evaluating their students work? RQ2: Do Software Engineering students take textual code review and inspection into account as a practice to improve software quality? RQ3: How to improve traditional teaching techniques used in Software Engineering students in order to improve the readability of written code? RQ4: How to get students to better understand the coding standards used to improve software quality? 3.3 Data Collection A survey composed of 10 questions was utilized. It was applied to 58 students from the first and second semester of the Software Engineering Bachelor Program during the summer of 2020.

4 Results Of the 58 respondents, 40 gave the highest score to the query regarding the importance of applying programming standards. While 16 students only gave him a high score and 2 an average score. While it is true that more than 95% of respondents agree with the relevance of standards, 3.4% of them apparently do not think so. Given that respondents are currently first and second graders, i.e. they are having their first approach to the programming subject, it would seem normal for 3% of them to think that standards are only moderately relevant. Nevertheless, this little percentage represent students who are not fully aware of the importance of the use of coding standards. Furthermore, only 50 of the 58 respondents answered they apply programming standards regularly, while the remaining 8 apply them less frequently. It should be noted that there are external factors such as pressure, stress, lack of time, which indirectly influence the student’s decision if it is worth worrying about writing readable code. Hence, it is common to find students who believe coding standards are important but do not apply them all the time. One case in particular is group projects where 88% of students think that the use of standards significantly facilitates teamwork, while 10.3% do not think it has such a relevant impact and only 1.7% rate the use of standards as unnecessary. The 75.9% of students believe they would understand their code if they looked at it two months after writing it, while 24.1% think they probably could not do it. In other

50

D. Avila et al.

words, students who apply standards have the confidence to be able to understand and probably reuse their code at some point in the future. Similarly, it was found that 94% of students have had at least once difficulty understanding the code of a third party, whether this is a classmate or code downloaded from the internet. The role of the educator is important in the learning of standards, a fact evidenced by 89.7% of students who opine it is much easier to develop good programming practices when the professor teaches them and consistently applies them in classes. In addition, 87.9% of students agree with the premise of using code quality as a metric when grading assignments or projects, a fact that clearly encourages the writing of readable code. The 96.6% believe that poorly structured code with lack of programming standards will cause problems in the future, because when they want to modify or maintain, it will make it very difficult to understand again what they already had done. Most students surveyed are aware of the importance of learning standards and good programming practices, but it can be noticed that 17.2% do not consider them important or consider them boring, remarking that many do not give the relevance that this matter deserves. Finally, 86% of students are aware that the good programming habits they learn during their academic training can make a difference when he or she wants to enter the labor market.

5 Discussion In our study, participants who are considered novice developers have difficulty understanding other people’s code. This is attributed to the fact that most programmers write functional code but not code that has the necessary programming metrics and standards, code that besides improving the quality of a project would also minimize time and resources consumption in the future. Good coding practices are an essential part of the introductory classes of software engineering [6], and the best place to prevent the writing of unreadable code is on university while the developer is still going through academic training. So that illegible code should never be an issue on his professional life [8].

6 Conclusions Most students acknowledge the importance of programming standards, and their usefulness in software development. Despite this and due to external factors, such as pressure, stress, lack of time, they cannot always apply them in their academic activities. The readability of the code is acquired according to the experience and learning of each developer, so it is of great importance the study of standards, from the first university courses. Students are aware that, in order to have optimal teamwork when working on projects, a proper management of coding standards is needed because the understanding of each developer’s code is bind to their use.

Unreadable Code in Novice Developers

51

References 1. Dos Santos, R.M., Gerosa, M.A.: Impacts of coding practices on readability. In: Proceedings - International Conference on Software Engineering, pp. 277–285. IEEE Computer Society, New York, New York, USA (2018) 2. Foucault, M., Blanc, X., Falleri, J.R., Storey, M.A.: Fostering good coding practices through individual feedback and gamification: an industrial case study. Empir. Softw. Eng. 24, 3731– 3754 (2019) 3. Palomba, F., Bavota, G., Di Penta, M., Oliveto, R., De Lucia, A.: Do they really smell bad? a study on developers’ perception of bad code smells. In: Proceedings - 30th International Conference on Software Maintenance and Evolution, ICSME 2014, pp. 101–110. Institute of Electrical and Electronics Engineers Inc. (2014) 4. Martin, R.C.: Clean Code : A Handbook of Agile Software Craftsmanship (2009) 5. Sampaio, I.B., Barbosa, L.: Software readability practices and the importance of their teaching. In: 2016 7th International Conference on Information and Communication Systems, ICICS 2016, pp. 304–309. Institute of Electrical and Electronics Engineers Inc. (2016) 6. Sripada, S.K., Reddy, Y.R.: Code Comprehension Activities in Undergraduate Software Engineering Course - A Case Study, December 28 (2015) 7. Neto, V.L., Coelho, R., Leite, L., Guerrero, D.S., Mendonca, A.P.: POPT: a problem-oriented programming and testing approach for novice students. In: Proceedings - International Conference on Software Engineering, pp. 1099–1108 (2013) 8. Wang, Y.Q., Xu, X.F., Lei, L., Wang, J.Z.: An AHP-based evaluation index system of coding standards. Proc.- Int. Conf. Comput. Sci. Soft. Eng. CSSE 2008, 620–623 (2008) 9. Runeson, P., Höst, M.: Guidelines for conducting and reporting case study research in software engineering. Empir. Softw. Eng. 14, 131–164 (2009) 10. Yin, R.K.: Case study research. Design Methods Sage 4, 264–267 (2009)

Security Smart Farming System – AgriPro Janani Tharmaseelan(B)

, S. V. Aaran, D. Ravipprasath, and P. Jayapragash

Srilanka Institute of Information Technology, Malabe, Sri Lanka [email protected], {it17159272,it16124318, IT17143882}@my.sliit.lk

Abstract. In Sri Lanka many problems have to be dealt with, to be a successful farmer. Poor weather prediction, crop-raiding by wildlife animals and poor agricultural management are some of the main problems the farmers have been dealing with for a long time [1]. We were looking forward to providing a solution for the farmers on the above-mentioned issues. Currently, Farmers cultivate crops based on their past experience. This is turning out to be difficult because of the changes in weather patterns due to industrialization and global warming. We realized this could be overcome by an application that could process the data present in the agricultural and meteorological departments and give farmers insights about weather changes. Crop-raiding is another problem that farmers need to face in Sri Lanka. Farmers use traditional methods as fencing to prevent these attacks. But still, the wildlife intrusions happen over a broken fence [2]. The time taken to notify farmers regarding these problems ensures that the damage has already been done by the time they are informed. As a solution, we are proposing to implement a system to identify the animals and notify the farmer as soon as possible. Poor agricultural management implies a lack of knowledge about diseases, agrochemicals, and poor planning on cultivation [3]. We are planning to improve this situation by gathering information regarding these and provide that information to farmers collectively through an application. The final output of our research would be an application including a weather prediction, wildlife intrusion detection, and a sentimental analytical chatbot together Keywords: Chatbot · Artificial Intelligence · Machine learning · Smart farming

1 Introduction Innovation is more important in modern agriculture than ever before. The industry, in general, is confronting huge difficulties, from increasing expenses of supplies, a shortage of neighbors, and changes in consumer inclinations for transparency and manageability [4]. Agricultural technology is among the most revolutionary and progressive areas of modern technology, driven by the fundamental need for food [5]. Nowadays farmers in Sri Lanka plan their agricultural activities considering their experience on weather and season. But due to the changes in the weather patterns caused by industrialization, global warming, and deforestation, it is getting hard to predict the weather and plan accordingly [6]. The cost equation for having access to reliable weather forecast information is not © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 52–62, 2021. https://doi.org/10.1007/978-3-030-72654-6_6

Security Smart Farming System – AgriPro

53

always easy to quantify [7]. Farmers frequently take small yet critical decisions regarding their crops, which would collectively make the financial risk taken very large. Irrigation planning is a good example. If a farmer relies on a forecast for precipitation that turns out to be accurate, he saves the cost of unnecessary irrigation. And by having a good idea of the expected amount of rain over a period and irrigating just enough to allow crops to thrive, he will maximize yield. Until now in our country, there are no implemented techniques in usage, but the agriculture department and meteorological department keep some raw data on their website, but it is not helpful for farmers to overcome these problems. So, for farmers’ convenience, this research considers these factors and recommends our system AgriPro as a solution to them and hope this will be helpful for the farmers to achieve their expected profit from their cultivation and lead a happy life. It is generally accepted that conflicts between farmers and wildlife animals have always happened. “Crop raiding” refers to wild animals damaging plant crops cultivated by humans, by either feeding on or trampling, them. The severity of damage caused can be highly variable, depending on the animal and crop species involved and the value of affected crops to their owners [8]. Wildlife intrusion into the farming land is a problem faced by the farmers in Sri Lanka also. Farmers in Sri Lanka don’t know how to react to wild animal attacks on their crops because they aren’t receiving training on how to deal with wild animals raiding their crops, despite human-wildlife conflicts growing more frequently as farms expand into wildlife habitats, according to a study [9]. Traditionally wildlife invasions are prevented by building fences around the farming land. But still, the farming land is being invaded by some animals such as elephants, wild boars, etc. In order to solve this issue, it is needed to check the farmer’s issue in the mindset of a farmer. Farmer’s total income and their base of life come from crops. So, if an attack happens, he will not only lose his seasonal income but also his faith in life. Because for the farmers, they believe, their sustainably of life is their farm. That is one of the main reason for farmer’s suicide [10]. Even though the farms could be protected with the electrocuted fences, but the costs are so high it is unaffordable for the farmers, and this proves to be dangerous [11]. In addition to this problem, elephants are getting adapted to these electric fences [12]. It is turning out to be more clear that there is an undeniable need to understand indigenous or local knowledge systems in order to help adapt and promote technologies that will be suitable for local situations. If the farmers could get to know about the wildlife intrusion faster, they could act faster to take action regarding the wildlife intrusion. Lack of knowledge about agricultural technologies is another problem found among farmers [3]. Most of the farmers fail to keep up with the modern technologies used in the agricultural field. Failing to identify the diseases, not being able to use the agrochemicals and fertilizers correctly are some of the issues faced by the farmers due to lack of knowledge. This happens because in this ever-evolving world it is hard to keep up daily with the problems and solutions. It would be better if they had all the information in one place so that they could find it easily and get help easily.

54

J. Tharmaseelan et al.

2 Related Work There are many software applications available online which fulfill the basic requirements of the agricultural industry. But these systems do not meet the criteria to solve the impacts caused by environmental changes in real-time. Precautions or preventions are not provided when needed. And the applications are quite expensive, where it is not efficient for farmers of less income to purchase. The research system will fill the divergence for local farmers, helping them to access forecasts that will help take precautions and preventions on time. Agavance Grain is a similar application, which has a grain management system, utilizing the latest technology. This application also allows you to see information like commodity market values. This application provides real-time analysis of marketing activities. [13]. Crop Planning is another similar application from AgWorks, which acts as a complete agronomy crop planning tool. This application has features like sales management to provide sales activity information and could track insect refuge for GMO seeds. [14]. Crop Tracker is an operation management software application. It helps farmers to manage their products by providing tractability. Through its advanced data collection, analyzing, and reporting systems, Crip Tracker achieves operational efficiency. [15]. The table below shows the comparison between our application and similar applications available in the apple and google store and shows the uniqueness of our application by analyzing the features respective applications pose (Table 1). Table 1. Comparison with the available systems. Application

Real-time analysis

Onsite equipment support

Agvance Grain [13]

✓

✓

✓

✓

✓

AgOS® Crop Planning [14]

Prevention deliverables

✓

Crop Tracker [15] AgriPro

Chatbot providing relevant information

✓

✓

✓

✓

According to the table, it can be suggested AgriPro has incorporated all the components needed for the domain.

3 Methodology AgriPro is a real-time system to serve as a solution provider to mitigate and support agriculture and farmers through information technology. According to the research scope,

Security Smart Farming System – AgriPro

55

the requirement was to develop the system to overcome environmental issues for vast area farming with location-based precise results and to create a communication media to support and serve rural farmers. AgriPro is powered through two main parts, the frontend, and backend. The frontend is a mobile application that interacts with the user and backend processes as the backbone structure of the entire system, handling logic and integrations. The mobile application is implemented in React native a JavaScriptbased mobile development framework created by Facebook and is hosted and supported in both Android and IOS platforms. The backend is developed in Node.js which is a JavaScript run-time environment developed in Express js framework. This part of the paper contains the overall approaches and the technologies used in and implementation development. User interfaces, front end data storage, and API connectivity with backend are the three main development stages considered in frontend implementation. Overall the mobile application works in a cycle where user interfaces interact with users to provide, gather information, and send it through API gateways to the backend. The backend is created as the backbone of the system to handle the main logic and to integrate with third-party platforms for data gathering. The main tasks of the backend are data communication, manipulation, and algorithmic processing. The functionalities are developed through machine learning and artificial intelligence principles.

Fig. 1. System overview diagram

Figure 1 shows the System Overview Diagram of the project. The client mobile application will be connecting to the server through API. Internal and cloud storage will be used for the manipulation of efficiency. Machine learning processes will use cloud storage to minimize storage usage. User related processes are connected to MySQL

56

J. Tharmaseelan et al.

database. Server will generate results through prediction algorithms and logic modules to find out the exact outcome. Adam optimization algorithm is one of the main machine learning algorithm used in our project. Adam is an optimization algorithm which can be used to update network weights iterative based in training data. Diederik Kingma from OpenAI and Jimmy Ba from the University of Toronto proposed the algorithm in their paper. Adam combines two extensions of stochastic gradient descent and provides the advantages of both. Adaptive Gradient algorithm and Root Mean Square Propagation are the two extensions combined in Adam optimization algorithm [16] This algorithm is mainly chosen because it is straightforward to use and also computationally efficient along with fewer memory requirements. Bellow sections will cover how authors have implemented each component and their user interfaces. 3.1 Environmental Prediction, Pre Planning and Management Agriculture has more impact on environmental changes, so the authors selected considered this factor to be included in the system. Farm pre-planning, environmental change prediction & precautions, irrigation management, and bush fire identification are the main functionalities to be implemented. Above mentioned functionalities were interconnected whereas it represents a farming cycle. The pre-planning functionality is expected to deliver a list of selected plants which is suitable to grow in a user-specific location under different environmental circumstances in a cultivation cycle through pre-planning based statistics and predictions that indicate future weather changes along with on-time notifications and precautious guidelines. The functionality requires information on user location-based weather pre-data and growth rates of plants for different climates. Once the user navigates to the respective component in a mobile application, it triggers an API call to the backend with the current location of the device. Based on the location, the backend system will perform machine learning algorithms using pre-data to find future changes in weather patterns and will calculate average weather statistics for each month. The system will then compare the average values with the plant growth rate based on the cultivation period and will send the results to the frontend. Weather prediction values calculated from the above function will be used to find the expected waterfall over an area. Based on the plant type, the user will be provided information on the required water level for irrigation management. The application also has the bush fire identification feature which is implemented through image processing techniques. Farmer location-based satellite images will be gathered and processed to find the pixel area of objects that overlaps and collides with another. Continuous monitoring will help to find the pixel area deviation and will be used to determine the disaster. Once found the user will be notified with the direction and the speed of spread. 3.2 Chatbot and After Sales Management In the recent time, chatbot has become one of the most commonly used AI software, which has been adapted in most of the industries to simulate chat with the user by using natural language via any messaging application, websites. Evaluating and comparing

Security Smart Farming System – AgriPro

57

the existing chatbot systems in terms of efficiency, effectiveness, and user satisfaction is a great challenge in order to define a new solution for the agriculture industry. This chatbot architecture is categorized mainly into three sub-processes such as text processing, semantic understanding, and response generation. Moreover, to comply with the trigger response approach, knowledge should be segregated into’units’. Each of these units signifies a single question, and a distinctive combination of functions & parameters produces a certain response. In order for the conversion service to identify which response should be sent to the user, it shall calculate the similarity between the user input and all trained queries. This is performed by machine learning algorithms using similarity measures between sentences in which each word represents a single dimension. To improve accuracy, the model is trained by feeding thousands of correctly labeled examples of user inputs. Sample queries: • What crops could be grown in this climate? • What are the new crop diseases?

Fig. 2. Chatbot functionality

Figure 2 shows the mechanism of the chatbot. When a user inputs a query, the chatbot will understand the query, search within its knowledge base, and give the user the answer to their query as an output. Furthermore, sentiment analysis will be conducted while the conversation occurs, to understand and collect the feedback of the user.

58

J. Tharmaseelan et al.

When a sentence is detected, first the punctuations will be removed. Then the connecting words will be removed. Neutral words in the remaining sentence will be removed and with the remaining words, the value of the sentence would be calculated. Positive words like good would yield a positive value and negative words like bad would yield a negative value. Then the overall value of the sentence will be calculated and the review would be decided as either positive or negative. 3.3 Wild Life Intrusion Detection This system provides high security and sends immediate alerts via the mobile application to the owner when wildlife intrusion is detected. It is very essential to conduct image processing hence detection of wildlife intrusion is identified before the disruption. This image will be captured and saved in the database to be shared with the user with a notification. The notification will be sent via the Internet or GSM to the user’s application. The camera’s second channel will provide live streaming. The captured image will be saved to a permanent database and the live broadcast will be saved to a temporary database. Figure 3 below shows us the diagram of the wildlife detection system.

Fig. 3. Wildlife detection mechanism

4 Results and Discussion The authors tested the application with the dataset collected manually. To extract the features from the datasets different machine learning algorithms were used. To forecast weather Linear Regression is used [17]. Sentimental Analysis is used for the applications chatbot [18]. A teachable machine is used to train the application to detect wildlife animals [19]. The accuracy of the developed model is increased through the given training.

Security Smart Farming System – AgriPro

59

The application was able to predict and warn with an accuracy of 92 %. The important finding of the study is based on algorithm calculations. The prediction and precaution services provide statistical values of future changes which are useful for proper farm management. AgriPro was able to find suitable crops to farm in a given location. It predicts future weather patterns and environmental changes on a long term analysis. AgriPro would notify the users regarding the required level of water supply needed for the period based on its weather predictions. In any case of bushfire, the AgriPro would detect it in its initial stage and notify users regarding the incident. The crop-raid by wildlife animals will also be detected by the AgriPro and it would notify the farmers regarding the intrusion. User feedback was gathered through sentimental analysis and AgriPro includes a public chat ability. It will assist farmers through the sentimental analysis and farmers would be able to chat and find solutions among the farming community.

Fig. 4. Main dashboard

Figure 4 shows us the interface of the application where the functionalities of environmental prediction, pre planning, and management could be accessed. Figure 5 shows an interface the application will be showing when an animal intrusion happens in the farm or its surrounding area. Figure 6 represents a sample chat between a user and the chatbot. Users would be able to converse with the chatbot about the diseases and respective remedies. Chatbot responds to the user by identifying the keywords in the query of the user and provide the answer related to the query.

60

J. Tharmaseelan et al.

Fig. 5. Intrusion notification

Fig. 6. Chat bot

Security Smart Farming System – AgriPro

61

5 Conclusion This research was intended to provide solutions through modern IT methods and standards to solve the above-related issues and to make it reach low-income farmers and create economic equality among them. Functionalities will provide farmers a statistical analysis for future environmental changes and precautions to mitigate them. Further study revealed that irrigation management and natural disaster tracking can also be implemented through the ideology that has been followed. Usage of machine learning and algorithmic processing have paved the way to reduce the usage of onsite devices and management which is a unique feature of the research compared to other available applications. Each country invests and improves the quality of agriculture to resolve food problems and food insecurity. Rural farmers full fill the basic productivity requirement in developing countries, whereas large scale organizations are impacting the wealthy economies. It is simply decidable that agriculture is needed to fulfill basic human needs and is a pillar of a country’s economy. In order to address the requirement level, the farming industry should produce a healthy outcome to serve customers and to have a good reputation as a company in the compatible global market. Therefore, it is important to maintain farming through proper methodologies to control and mitigate unexpected occurrences under different circumstances regardless of the organization’s economic background or manpower contribution. Larger companies follow information technology-based methodologies to manage farms and find out statistical data through on-site IoT devices. They process them and plan the future with better ways of solutions. But rural farmers are not capable of consuming such products due to the cost, thus they follow traditional methodologies and human instincts and experiences to overcome. Although there are technologies and acceptable methodologies for farm maintenance, they are based on the legacy system, which incase can be considered as less information technology impacted. Maintenance.io is a CMMS software solution that helps farmers manage the maintenance of their equipment [20]. With the current inventions and improvements in the ICT field; it is much possible to provide services with less hardware depended and effective to overcome future needs through statistical analysis. This can be helpful to modify the existing technologies and make them reach low-income farmers. As future work, It would be better if the intrusion can be detected before some distance. That would be more beneficial for the farmers to safeguard their farms. Weight Sensors can be used, before some distance, and from that identifying the intrusion can be considered to be implemented in the future. Acknowledgment. We would like to extend our heartfelt gratitude towards those who have supported us with this research and in many ways. Last but not least we would like to take this opportunity to thank Sri Lanka Institute of Information Technology (SLIIT) for lending us a suitable environment. We are grateful immensely to all who supported and guided us throughout this research.

62

J. Tharmaseelan et al.

References 1. Federation of Chambers of Commerce and Industry of Sri Lanka. https://www.fccisl.lk/upd ates/10-main-issues-faced-by-the-farmers-and-agri-based-entrepreneurs-in-uva-summedup-at-the-kpf. Accessed 14 Sep 2020 2. Mongabay. https://news.mongabay.com/2017/06/breaking-a-fence-breaking-habit-mainta ining-the-fences-that-reduce-human-elephant-conflict. Accessed 14 Sep 2020 3. Pressreader. https://www.pressreader.com/sri-lanka/daily-mirror-sri-lanka. Accessed 24 Sep 2020 4. PLUGANDPLAY. https://www.plugandplaytechcenter.com. Accessed 12 Sep 2020 5. New World Encyclopedia. https://www.newworldencyclopedia.org/. Accessed 12 Sep 2020 6. The Sunday Morning. https://www.themorning.lk. Accessed 12 Sep 2020 7. National Weather Service. https://www.weather.gov. Accessed 12 Sep 2020 8. Hill, C.M.: The International Encyclopedia of Primatology. Oxford Brooks University, United Kingdom (2017) 9. Mongabay. https://news.mongabay.com/. Accessed 12 Sep 2020 10. Behere, P.B., Bhise, M.C.: Farmers’ suicide: across culture. Indian J. Psychiatry 51(4), 242– 243 (2009) 11. Big Life Foundation. https://biglife.org/where-we-work/area-of-operation/crop-raiding. Accessed 12 Sep 2020 12. Newswire. https://www.newswire.lk. Accessed 12 Sep 2020 13. Agvance. https://www.agvance.net. Accessed 10 Sep 2020 14. Agworks. https://www.agworks.net/planning. Accessed 10 Sep 2020 15. Crop Tracker. https://www.farms.com/agriculture-apps/spraying/crop-tracker-dragonfly. Accessed 10 Sep 2020 16. Kingma, D.P., Ba, J.: Adam: a method for stochastic optimization. In: ICLR (2015) 17. Bendre, M.R., Thool, R.C., Thool, V.R.: Big data in precision agriculture: weather forecasting for future farm-ing. In: 1st International Conference on Next Generation Computing Technologies (NGCT), Dehradun, pp. 744–750 (2015) 18. Medhat, W., Hassan, A., Korashy, H.: Sentiment analysis algorithms and applications: a survey. Ain Shams Eng. J. 5, 10931113 (2014) 19. Teachable Machine. https://teachablemachine.withgoogle.com/. Accessed 17 Sep 2020 20. Maintenance. https://maintenance.io/industries/agriculture-and-farming-maintenance-man agement-software/. Accessed 14 Sep 2020 21. Pressreader. https://www.pressreader.com/. Accessed 12 Sep 2020

Smart Contract Code Generation from Platform Specific Model for Hyperledger Go Mantas Jurgelaitis(B)

ˇ , Vaidotas Drungilas, Lina Ceponien˙ e , Evaldas Vaiˇciukynas, ˇ Rita Butkien˙e, and Jonas Ceponis

Informatics Faculty, Kaunas University of Technology, Kaunas, Lithuania [email protected]

Abstract. Model Driven Architecture (MDA) together with Unified Modelling Language (UML) presents a framework which transfers the emphasis of development from source code to the higher level of abstraction i.e., models. In this paper, we demonstrate the application of MDA principles for generating smart contract code executed on a blockchain. Even though blockchain smart contracts are not in all cases a classic type of object-oriented software, which UML is intended for, we demonstrate the possibility to adapt to this specific implementation platform. MDA Platform Specific Model (PSM) is used as an input for transformation algorithm which maps PSM metamodel elements to Go Chaincode elements and produces Go chaincode. In PSM, UML class and sequence diagrams are used for specifying structural and behavioural aspects of the smart contract. MOFM2T transformation language and Acceleo tool are employed for the implementation of this algorithm. The results of the algorithm execution were demonstrated using example chaincode for machine learning model validation. Keywords: MDA · Blockchain · Chaincode · PSM · Code generation

1 Introduction Over recent years, blockchain technologies matured enough to be used in more use cases than just cryptocurrencies and various tokens. The number of areas of blockchain technology application is constantly increasing as it is being employed in the sectors of finance, education, energy, medicine, real estate, government, scientific research, gaming, etc. [1, 2]. Blockchain itself is a distributed ledger based on principles of decentralization and immutability, but its utilization can be expanded by introducing the decentralized programs working on a blockchain network. These programs are called smart contracts. The implementation of a smart contract logic is developed using a built-in scripting language or adapted available programming languages. While not all blockchains support smart contracts, amongst the most popular ones are Ethereum and Hyperledger [1]. In both platforms, the additional business logic triggered by a transaction is executed by a virtual machine hosted on the respective blockchain. Hyperledger is a blockchain ledger technology stack providing open-source enterprise-grade blockchain deployments solution. One of such solutions is Hyperledger © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 63–73, 2021. https://doi.org/10.1007/978-3-030-72654-6_7

64

M. Jurgelaitis et al.

Fabric with an additional layer of access control, that can be used to manage additional organizational, role permissions [3]. The smart contract, or chaincode as it is called in the Hyperledger Fabric ecosystems, can be written in a few different programming languages, such as Go, JavaScript, or Java. Although the area of application of blockchains and smart contracts is broad, the development of blockchain based systems is still a challenging task. The blockchain platforms are evolving, so do the smart contract implementation languages. The developer not only has to know certain platform specific implementation details but also must deal with changing platform specific features, revealed security issues and even situations when a smart contract developed for the previous version of a blockchain platform cannot be run on the newer version. These problems could be reduced using the modelling principles elaborated in Model Driven Development (MDD) approaches. Model Driven Development is propagating the idea, that modelling can at least relieve or even completely replace the work required for writing the source code [4]. A model can be used not only for visualization, communication, and documentation purposes but also for code generation. Model Driven Architecture (MDA) goes even further and defines a framework which encompasses a set of guidelines for specifying and employing models and their transformations in several steps to generate the source code of system under development [5]. MDA provides a clearly outlined framework [6], which could be adapted to many different domains, including blockchain. MDA is most used in tandem with Unified Modelling Language (UML). UML is an ISO adopted modelling language which encompasses several types of diagrams for representing the structure and behaviour of the software system [7]. MDA defines three models at different levels of abstraction: Computation Independent Model (CIM) for gathering initial business requirements, Platform Independent Model (PIM) for modelling the design of the system under development, but without implementation details, and Platform Specific Model (PSM) for detailed specification based on chosen implementation platform which can be transformed into the source code. The goal of our research is to demonstrate that the principles of MDA and UML diagrams can be successfully applied for the development of smart contracts for blockchain. In this paper, we present the algorithm for transformation from MDA PSM into Go code of the Hyperledger Fabric smart contract (chaincode). The developed algorithm is a part of a broader methodology under development which will encompass the whole MDA process of transformations from CIM to PIM, then to PSM and lastly to source code for a chosen blockchain platform [8]. The generated code was successfully deployed on a privately hosted Hyperledger Fabric blockchain. The paper is structured as follows. In the second section, related work overview is presented. In the third section, a proposal is outlined, algorithm and code generation templates for generating Go chaincode are presented. The fourth section presents the application of the algorithm implementation to a specific chaincode application for machine learning model validation. Lastly, conclusions and future work are presented.

2 Related Work Modelling is often used as a tool for problem domain analysis, documentation, visualization or even code generation. Blockchain could also benefit from the application of

Smart Contract Code Generation from Platform Specific Model

65

modelling approaches, during which specified models can be used as a basis for visual programming, model execution, or even executable code generation [9]. Several proposals were developed using MDD or MDA for the blockchain application design [10–12]. Most publications regarding code generation using MDA show that technology can be applied successfully to enhance software development [12]. Although blockchain is a relatively new technology, various proposals on applying models for blockchain have cropped up recently, ranging from model driven engineering [10], MDA [13], specification languages [14, 15], code generation [13, 16, 17] to execution engines based on business process modelling [11]. B-MERODE approach [10] is proposed for model driven engineering of business processes to generate blockchain based information systems. Since the proposed approach is based on existing MERODE method, by adapting it to the blockchain, it reuses layers and transformations defined in MERODE, and adds two additional layers – Core ISS Layer, Permissions Layer, to outline the specifics of blockchain based systems. In [13] authors present an MDA based approach for implementing a blockchain solution by focusing on the design of a blockchain application rather than the complicated technical implementation of the selected blockchain technology. Using the approach, the source code and deployment scripts can be generated for target platforms. UML and Smart Contract Ontology are used for PIM definition. Once the PIM-to-PSM transformation is executed, the specific code of the target platform is generated by using Acceleo M2T (Model to Text) transformation. The approach can generate both Ethereum and Hyperledger Fabric smart contract boilerplate source code and deployment scripts for the specific platform. Caterpillar [11] is a Business Process Management System that has an execution engine which transforms specified Business Process Model and Notation process models into running Ethereum’s Solidity smart contracts code. The generated artefacts code can be compiled and deployed on the Ethereum network. The authors of [14] propose a high-level object-oriented domain specific language called Contract Modeling Language (CML) that generates smart contract code through the automated application of design patterns. The basis of CML is a contract, similar to a class, it is used to specify constructs like Party, Asset, Transaction, Event, Function constraints. The constructs based on mapping rules generate Solidity code that relies on static and dynamically created support libraries. In [17] a Petri Nets based Secure Smart Contract Generation Framework is presented. Its implementation includes a graphical modelling tool for Petri Net workflow specification, a Petri Net simulation engine for validation, and verification and a Solidity smart contract template generating transformation engine. Authors focus on the security of the generated smart contract since the verification and validation can prevent workflows deadlocks or avoid potential logical business process errors. While using this approach developers generate the smart contract template, that still needs to be extended with additional business-logic before the deployment. The research in [16] focuses on the deployment view and transformations of the distributed ledger’s deployment model into Gradle Groovy deployment scripts for Corda network. The deployment model is modelled using UML and UML extension profile in using a variation of the 4 + 1 architectural view model for distributed ledger solutions.

66

M. Jurgelaitis et al.

Using the information specified in Deployment View, authors automate the generation of distributed ledger network configurations. Most of the covered research depends on model driven engineering techniques for business process analysis or even the production of implementation artefacts. Though no common unified approach to model business processes exists to support decentralization on the blockchain, it is obvious that the introduction of different abstraction levels would be of benefit [10]. Our approach is based on ideas of code generation from models presented in [11, 16, 17] and most similar in notion to the [13] as the authors generate boilerplate source code for the Hyperledger Fabric chaincode or Ethereum smart contracts. But in contrast to generating only boilerplate code, comprising names, fields of structures and functions with parameters, our research aims to generate the execution logic of methods. Our proposed transformation algorithm is a part of broader research on applying MDA principles in blockchain development for business and scientific processes [8]. To demonstrate the possibilities of blockchain application in scientific processes we use an example of chaincode for machine learning model validation to demonstrate the code generation algorithm execution results.

3 Algorithm for Transformation from PSM to Go Source Code Our proposal is based on the principles of MDA and in detail describes the process of transforming PSM for Hyperledger Fabric Go chaincode into Go source code. We use behaviour information from UML sequence diagrams to generate not only the chaincode structure source code but also the logic of the functions. The proposed algorithm maps PSM metamodel elements to Go language metamodel elements to produce Go chaincode. The main elements of PSM metamodel are outlined in Fig. 1. And Go language metamodel is presented in Fig. 2. The mappings between metamodels are also outlined using stereotypes in PSM metamodel. In our solution, PSM for Go encompasses UML class and sequence diagrams. Class diagrams represent the structure and sequence diagrams specify behaviour details of the smart contract under implementation. UML Class metaclass (Fig. 1) is used as a basis for chaincode code generation. Therefore, Class has the stereotype «Chaincode» and is mapped to Chaincode metaclass in Go metamodel (Fig. 2). Each Go Chaincode can import Packages (Go chaincode must import at least one shim API package [18], used to communicate to the specific ledger instances). The Class in PSM metamodel can have SubClasses (which are mapped to Structure metaclasses in Go metamodel), also each Class can have its Properties (Fields in Go metamodel). Additionally, each Class can have defined Operations, which in turn have their Parameters (Functions and their Parameters in Go). Each of the Parameters can have different types and names. Lastly, Operation as a Behavioural Feature can be defined by a behaviour model, in this case by Interaction. Each specified Interaction metaclass denotes a specific Operation implementation and is used to generate the Go function execution logic code. Each Interaction in PSM metamodel is made up of several InteractionFragments. Each of these fragments is either a CombinedFragment or OccurrenceSpecification.

Smart Contract Code Generation from Platform Specific Model

67

Fig. 1. Fragment of PSM metamodel for Hyperledger Fabric Chaincode in Go.

Fig. 2. Hyperledger Fabric Go chaincode metamodel.

CombinedFragment can have different InteractionOperators, based on which Go Statements for the Go Function metaclass are generated. Other Go Expressions like variable declarations, function calls are generated based on specific PSM Message types denoted in each InteractionFragment. The main part of the Go code generation algorithm is presented in Fig. 3. Code generation starts with chaincode file creation, which is appended with the code line by line while reading the data specified in the PSM model. If Interaction in PSM for a specific Operation exists, the Interaction is transformed into a Function’s implementation code. The Interaction transformation algorithm is detailed in Fig. 4.

68

M. Jurgelaitis et al.

Fig. 3. Main algorithm for generating Go chaincode.

Fig. 4. Algorithm fragment for transforming InteractionFragment set into the Go Function source code.

The interaction transformation starts by reading the Interaction and the ordered set of InteractionFragments. After determining CombinedFragment InteractionOperator kind, an alternative or loop statement is appended to the code. As each CombinedFragment can hierarchically nest different InteractionFragments, a transformation of the InteractionFragment set is performed recursively. The presented algorithm is implemented using Eclipse Acceleo plugin which imports the PSM model as XMI file. A developed Java solution accepts an XMI format UML model and applies transformations in MOFM2T to output a Go chaincode file. For this purpose, the MOFM2T templates were implemented using Obeo Acceleo plugin and several libraries provided with the plugin. The code is generated under few assumptions: that in PSM the messages, that correspond to an operation, will have either all arguments specified or none; the combined fragments will have the operand denoted (in a loop either the operand, or a minInt); all properties, parameters will have a specified type; call messages will have at least a name; create messages will be either empty or have all literals specified; return message will be connected to gate. The MOFM2T template for implementation of algorithm part for chaincode structure generation is presented in Fig. 5. This template maps PSM class diagram elements into Go chaincode structure.

Smart Contract Code Generation from Platform Specific Model

69

[template public transformModel2Chaincode(model : Model)][file (model.name.concat('.go'), false)] package main import "github.com/hyperledger/fabric-chaincode-go/shim" [for (e : Element | model.ownedElement)][let c : Class = e] [for (cn : Class | c.ownedElement->filter(Class))] type [cn.name/] struct{ [for (p : Property | cn.attribute)] [p.name/][p.type.name/] [/for]}[/for] [for (o : Operation | c.ownedOperation)] func [o.name/] ([for (pr : Parameter | o.ownedParameter) separator(', ') ?(not(pr.direction = ParameterDirectionKind::return))][pr.name/] [pr.type.name/][/for]) ([for (pr : Parameter | o.ownedParameter) separator(', ') ?((pr.direction = ParameterDirectionKind::return))][pr.type.name/][/for]) {[for (b : Behavior | o.method)] [let it : Interaction = b] [transformInteraction2Function(it.fragment)/][/let][/for]} [/for][/let][/for] [/file][/template]

Fig. 5. MOFM2T template for implementing the code generation algorithm: chaincode structure generation.

The other two presented MOFM2T templates (Fig. 6) are used for mapping PSM sequence diagram elements into chaincode functions code. These fragments of transformation implementation in MOFM2T demonstrate transformation of typical code elements – loops and conditional statements. Alternatives fragment in PSM sequence diagram is mapped to if statement in Go (altf template). Loop in PSM sequence diagram is mapped into for statement in Go (loopf template). [template public altf(operand : InteractionOperand)] [if (operand.precedingSiblings()>isEmpty())]if([operand.guard.specification.stringValue()/]){[elseif (operand.followingSiblings()->isEmpty())]else{[else]else if([operand.guard.specification.stringValue()/]){[/if][/template] [template public loopf(operand : InteractionOperand, int : Integer)] [if (operand.guard.specification->notEmpty())]for ([operand.guard.specification.stringValue()/]) [elseif (operand.guard.minint->notEmpty())]for ['i'+int/] := 0; ['i'+int/] 60%, is the same as in the null hypothesis

H.1.20 , π2 < = 60%, π2 is the perception percentage that evaluates the ease of understanding of the approach examples

H.1.2, π2 > 60%, π2 is the same as in the null hypothesis

H.2.10 , π3 < = 60%, π3 is the perception percentage that evaluates that the approach has enough information for its application

H.2.1, π3 > 60%, π3 is the same as in the null hypothesis

H.3.10 , π4 < = 60%, π4 is the H.3.1 π4 > 60%, π4 is the same as in the null measurement that evaluates the approach hypothesis utility perception to build a scope that guides and delimits the process line H.3.20 π6 < = 60%, π6 is the H.3.2. π5 > 60%, π5 is the same as in the null measurement that evaluates the perception hypothesis that the approach is organized and consistent H.4.10 , π5 < = 60%, π5 is the measurement that evaluates the utility perception of the obtained result for a process line development

H.4.1. π6 > 60%, π6 is the same as in the null hypothesis

Table 3. Summary of calculations for hypothesis validation Hypothesis Z (statistical) Vc

Hypothesis testing

H.1.10

−1.035

1.28 It is not rejected

H.1.20

−1.035

1.28 It is not rejected

H.20

1.380

1.28 Is rejected

H.30

2

1.28 Is rejected

H.3.20

1.370

1.28 Is rejected

H.4.10

1.723

1.28 Is rejected

• H.20 is rejected and, the alternative hypothesis H.2. is accepted. In this way, users of our proposal perceive that the approach has enough support information to guide its application. • H.3.10 is rejected and, the alternative hypothesis H.3.1 is accepted. In this way, the users of our approach perceive that the approach is useful for building a scope that helps in the process line delimitation.

A Scoping Definition Experiment in Software Process Lines

81

• H.3.20 is rejected and, the alternative hypothesis H.3.2 is accepted. In this way, the users of our approach perceive that the approach is organized and consistent. • H.4.10 is rejected and, the alternative hypothesis H.4.1 is accepted. In this way, users of our approach perceive that the result of executing the approach is useful for the process line development. Quantitative Analysis and Results. The analysis was based on the suitability data comparison provided by the groups during the execution of our approach, in order to establish whether they are statistically similar. The hypotheses for each of the process variable activities are shown below in Table 4: Table 4. Quantitative hypothesis Null hypothesis:

Alternative hypotheses

H.5.10 the suitability values by each group for the activity: plan the integration is similar

H.5.1 the suitability values by each group for the activity: plan the integration is not similar

H.5.20 the suitability values by each group for the activity: peer programming is similar

H.5.2 the suitability values by each group for the activity: peer programming is not similar

H.5.30 the suitability values by each group for H.5.3 the suitability values by each group for the activity: individual programming is similar the activity: individual programming is not similar H.5.40 the suitability values by each group for the activity: code generation is similar

H.5.4 the suitability values by each group for the activity: code generation is not similar

H.5.50 the suitability values by each group for H.5.5 the suitability values by each group for the activity: integrate each subsystem is similar the activity: integrate each subsystem is not similar

The Chi-Square test statistic (χ2) was used to make the analysis using the following Eq. (3): χ2 =

c r   (Aij − Eij)2 i=1 j=1

Eij

(3)

Where, Aij = is the value observed in the ith row, jth column, Eij = is the expected value in the ith row, jth column, r = rows number and c = columns number. Table 5 shows the observed and expected values in the specific case of the variable activity plan the integration. The first row refers to each of the groups that participated in the experimentation. The second row to the suitability value defined by each of the groups for the task Plan the Integration. The third row to how many of the group members agreed with the suitability value defined in row 2, and the expected value reflected in the fourth row. The χ2 calculation was done by using the highlighted rows data in Table 6 using Eq. (3), where χ2 = 6.39. The degrees of freedom for the χ2 test is given by (rows -1) * (columns

82

P. H. Ruiz et al. Table 5. Observed values of suitability by group to plan integration

Group

1

2

3

4

5

6

7

8

9

10

11

12

Suitability per group 0,58 0,58 0,62 0,58 0,63 0,66 0,38 0,60 0,50 0,62 0,52 0,60 Observed values

3

3

3

3

3

3

2

3

3

3

3

3

Expected value

2

2

2

2

2

2

3

2

2

2

2

2

-1), which for this case is 11. The critical value is established by using the distribution table of χ2 with 11 degrees of freedom and significance α = 0.5, therefore, vc = 19.67. Table 6 shows the result of χ2 for all the process variable activities. Table 6. Summary of calculations for hypothesis validation using χ2 Variable activities of the process

χ2

Vc

Hypothesis testing

19.67

Acceptance of H.5.10

Plan the integration

6.39

Peer programming

14.04

19.67

Acceptance of H.5.20

5.48

19.67

Acceptance of H.5.30

Code generation

13.86

19.67

Acceptance of H.5.40

Integrate each subsystem

15.12

19.67

Acceptance of H.5.50

Individual programming

According to the results of Table 6, it can be inferred that: • H.5.10 can be accepted. It can be said that the suitability values established by each group for the variable activity plan the integration are statistically similar • H.5.20 can be accepted. It can be said that the suitability values established by each group for the variable activity Peer programming are statistically similar • H.5.30 can be accepted. It can be said that the suitability values established by each group for the variable activity Individual programming are statistically similar • H.5.40 can be accepted. It can be said that the suitability values established by each group for the variable activity Code generation are statistically similar • H.5.50 can be accepted. It can be said that the suitability values established by each group for the variable activity Integrate each subsystem are statistically similar Threats to Validity. Construct validity: in order to minimize the subjectivity in the instruments supporting the collection of study information, the instruments underwent several validations in which members of UC and UM participated. Another threat of this type is the incorporation and management of new conceptual and language elements in the study. In order to reduce it, an initial activity was dedicated in which participants were socialized and contextualized in details about it to achieve a concept unification and language. Internal validity: the method execution results may be conditioned due to the use of the same problem and types of participants in the scope definition. To

A Scoping Definition Experiment in Software Process Lines

83

mitigate it, the study considered that the problem resolution was made through the different groups. Another threat of this type may be the time for the study execution, having long sessions, participants in the final stages could perceive fatigue which can influence the results, to mitigate it, in the middle of experimentation, participants took a break without communication among them. External validity: due to the experiment was conducted in an academic environment, the results could not be generalized to an industry context. To mitigate it, participants with experience, in software development, in process engineering, and with no experience were considered to try to emulate the industry context. In addition, to carry out the experiment, all the necessary supplies were provided to execute the method which does not happen in the industry.

5 Conclusions This work validates the ease of use, utility, and reliability of an approach to scope definition in an SPrL. For this, an experiment was developed with students from three universities. According to the hypotheses validation H.1.1 and H1.2, it can be said that the participants failed to have a complete understanding of the instructions, guidelines, and examples that are part of the approach, in such a way, this allows us to identify future improvements in the description and elements that make it up. In accordance with the validation of H.2. it can be concluded that the participants perceived that the approach has enough information for its application, but it cannot be ensured that this information is adequate to achieve a satisfactory understanding of the information. Although H.1.1 and H1.2 are inconclusive, participants perceived that there is a utility in the approach to support the scope delimitation of an SPrL, through an organized and consistent approach, in accordance with the H.3.1 and H3.2 validations. In addition, the H.4.1 validation allows us to say that the results obtained when executing the approach are also utility. From the above it can be inferred that our proposal is perceived as useful, organized, consistent and has enough information. However, there is a conceptual complexity that does not allow an adequate understanding of all the work involved in scope defining. Regarding the approach reliability, it can be said that the resulting suitability values are statistically similar. These values were achieved systematically and independently of the group for each of the process elements, in this way it can be said that the participants arrived at the same solution. Therefore, hypotheses H.5.10, H.5.20, H.5.30, H.5.40, H.5.50, are accepted, so that, in this context, the approach can be considered systematic and can, therefore, help determine the inclusion or exclusion of process features for a specific situation. Although this study was carried out in an academic environment, this may be a limitation for the results obtained, taking into account that, for a real environment, there will not be the same support team used in the development of this study, and therefore these results may change. For this reason, as future work, it is necessary to submit the proposal towards validation in a software organization to compare the results, in addition to make the corresponding improvements to the approach regarding its conceptual clarity, adding a qualitative layer to the model.

References 1. Humphrey, W.S.: Managing the Software Process. Addison-Wesley Longman (1989)

84

P. H. Ruiz et al.

2. Hossein, S., Natsu, C.: Characterizing a software process maturity model for small organizations. ACM SIGICE Bull. 23, 2–11 (1997) 3. Ginsberg, M., Quinn, L.: Process tailoring and the software capability maturity model. (1994) 4. Jacobson, I., Booch, G., Rumbaugh, J.: Unified software development process. (1999) 5. Hanssen, G.K., Westerheim, H., Bjørnson, F.O.: Tailoring RUP to a defined project type: a case study. In: PROFES2005 Proceedings of the 6th International Conference on Product Focused Software Process Improvement, pp. 314–327 (2005) 6. Carvalho, D.D.D., Chagas, L.F.: Software process lines: a systematic literature review. In: International Conference on Software Process Improvement and Capability Determination, Vilnius, pp. 118–130 (2014) 7. Hurtado, J.A., Bastarrica, M.C.: Building software process lines with CASPER. In: International Conference on Software and Systems Processes. ICSSP, pp. 170-179 (2012) 8. Rombach, D.: Integrated software process and product lines. In: Unifying the Software Process Spectrum, Volume 3840/2006, pp. 83–90 (2005) ISSN: 0302–9743 9. Armbrust, O., Katahira, M., Yuko, M., Munch, J., Nakao, H., Ocampo, A.: Scoping software process lines. J. Softw. Process Improv. Pract. Examining Process Design Change 14(3), 181–197 (2009) 10. Ruiz, P.H., Camacho, C.M., Hurtado, J.A.: A comparative study for scoping a software process line. In: ICAI Workshops (ICAIW), pp. 1–6 (2018) 11. Washizaki, H.: Building software process line architectures from bottom up. In: ProductFocused Software Process Improvement, pp. 415–421 (2006) 12. Schmid, K.: A comprehensive product line scoping approach and its validation. In: Proceedings of the 24th International Conference on Software Engineering, pp. 593–603 (2002) 13. John, I., Knodel, J., Lehner, T., Muthig, D.: A practical guide to product line scoping. In: 10th International Software Product Line Conference (SPLC2006), pp. 3–12 (2006) 14. Armbrust, O., Katahira, M., Yuko, M., Jürgen, M., Haruka, N., Ocampo, A.: Scoping Software Process Models - Initial Concepts and Experience from Defining Space Standards. In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), pp. 395–406 (2008) 15. Aleixo, F.A., Freire, M., Alenca, D.: A comparative study of compositional and annotative modelling approaches for software process lines. In: 26th Brazilian Symposium on Software Engineering (2012) 16. Martinez-Ruiz, T., Garcia, F., Piattini, M., Munch, J.: Modelling software process variability: an empirical study. IET Software 5(2), 172–187 (2011) 17. Pazin, M.G., Geraldi, R.T., OliveiraJr, E.: Comparing SMartySPEM and vSPEM for modeling variability in software processes: a qualitative study. In: Proceedings of the 17th Brazilian Symposium on Software Quality (2018) 18. Ramírez Álvarez, Y.L., Lasso Velasco, S.P., Ruiz, P.H., Hurtado, J.A., Hoyos Carvajal, P.: Customizable software processes: an exploratory case study in a small organization. In: 13 Conferencia Colombiana de Computación, Cartagena (2018)

Best Practices for OutSystems Development and Its Influence on Test Automation Joana Salgueiro2 , Fernando Ribeiro1,2(B) , and José Metrôlho1,2 1 R&D Unit in Digital Services, Applications and Content, Castelo Branco, Portugal

[email protected] 2 Polytechnic Institute of Castelo Branco, Castelo Branco, Portugal

Abstract. Low-code development platforms are designed to accelerate software development with a minimum of hand coding. They have allowed professionals with distinct backgrounds to become software developers. This brings more professionals to IT areas, reskilling some of them from different areas of knowledge but, at the same time, it has also brought to the software development area many professionals without a strong background in this area. Although testing should be performed in all applications, regardless of the programming language or the platform used, this reinforces the need for testing low-code developed applications. In this work, we study the test automation process on the OutSystems low-code development platform. The focus is on Unit, Integration/API and System/End-to-End testing levels. Examples illustrate that the implementation of best practices during the development process can have a significant influence on the test automation process. However, it is important to know whether the workload associated with the implementation of the best practices can undermine the benefits of using a low-code platform and it is important to consider that not all test automation tools have the same behavior. Keywords: Low-code platforms · OutSystems · Software quality · Software testing · Test automation

1 Introduction A recent trend that has been noticed is the growing adoption of low-code development platforms (LCDP) by IT companies. The State of Application Development [1] refer that 41% of respondents said their organization was already using a low-code application platform and, a further 10% said they were about to start using one. This growing interest is also corroborated by the Low-Code Development Platform Market [2]. It reports that the global LCDP market size is projected to grow a rate of 28.1% during the 2020–2026 period. In fact, in recent years, these platforms have become popular and are now widely used in many companies around the world. Many of the reasons given for this growth are in line with the advantages that are usually associated with the use of LCDP: they comprise many of the same tools’ functionalities that developers and teams use to design, code, deploy and manage an application portfolio [3]; a significant part of the job can be done through a drag-and-drop interface and although developers may still need to do © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 85–95, 2021. https://doi.org/10.1007/978-3-030-72654-6_9

86

J. Salgueiro et al.

some coding this is just for specific tasks [4]; they are able to accelerate the delivery of new software and applications, allowing to update and deliver new features in shorter time periods [5]; they allow for apps to be built for multiple platforms simultaneously, and cross-platform support [5]; and data integration capabilities have already been built and can be customized easily [5]. However, and although low-code development is often associated with error-free development, this is not true. Additionally, the characteristics of these platforms have allowed professionals with distinct backgrounds, to become software developers. If, on the one hand, this is a positive fact, and brings more professionals to IT areas, reskilling some of them from different areas of knowledge and of greater employability difficulties, it is also true that the lack of specialized knowledge of some of them can lead to a high number of bugs in the developed software. In fact, testing the software developed on these platforms is important but it is necessary to analyze various test strategies and tools that best suit these platforms. In these cases, test automation is even more essential for maintaining the productivity and quality of the software. However, it is important to consider their effects on cost, quality, and time to market of the software product. In this work we study the test automation process on the LCDP OutSystems. In Sect. 2 we present a background about test automation on Low-Code Development Environments. In Sect. 3, we analyze some works that addressed test automation on low-code platforms. In Sect. 4, we analyze two test automation tools and their suitability for OutSystems applications, and, in Sect. 5, we analyze the impacts that best practices in OutSystems Development can have as facilitators in the test automation process and in the feasibility of using these testing tools.

2 Test Automation on Low-Code Development Environments Manual testing is often difficult and impractical. Manual testing can be time-consuming and vulnerable to inaccurate results. There are even many situations in which manual testing approaches may not be effective, especially in cases where it may be useful to repeat tests already carried out such as regression tests. In many situations, automated tests should be considered. However, to be effectively valid, test automation may entail some significant changes in the development process that, in some cases, may limit some of the benefits normally associated with LCDP. Automation, if not well adopted, can imply a significant increase in the development time or increase the need for coding. It is also important to consider that in Behavior Driven Development (BDD), not all test scenarios need to be automated as some may be difficult to automate or may be more suitable for manual approaches. It is, therefore, important to assess the situations and implications in which testing automation would be beneficial and choose the right testing tools. 2.1 Test Automation in Agile Agile methodologies are commonly used together with LCDP. In these methodologies there are three testing techniques that are frequently used for test automation:

Best Practices for OutSystems Development

87

Test-Driven Development (TDD); Acceptance Test-Driven Development (ATDD); and Behavior Driven Development (BDD). TDD is a technique used to develop code guided by automated test cases. It is also known as test first programming since test cases are written before the code. The test cases written are primarily unit level and typically using white-box techniques. These tests may also be written at the integration or system levels. One limitation with TDD is that if the developer misunderstands what the software is to do, the unit tests will also include the same misunderstandings, giving passing results even though the software is not working properly [6]. ATDD is based on the test first concept. It defines acceptance criteria and test cases for it early in the development process, in the confirmation phase of a story development process. ATDD is a collaborative approach which allows every stakeholder to understand how the software component must behave and what the product owner, tester, developer, and others need to do to insure this behavior. In ATDD, it is the acceptance tests that will verify the implemented user story [6]. BDD starts from a point of view that behaviors of software are easier for stakeholders to understand when taking part in test creation than the specific tests themselves. Tests are then based on the expected behaviors, and the developer runs these tests all the time as they develop the code. BDD promotes immediate test automation even more than ATDD [6]. In BDD, acceptance tests are often formulated in a structured language, referred to as the Gherkin language. In this case requirements are defined in a ‘Given – When –Then’ format and they become the acceptance test cases and also serve as the basis for test automation [7]. 2.2 Testing on the OutSystems Low-Code Development Platform OutSystems applications benefit from a continuous integrity validation that checks the impact of all changes in the application layers to ensure that nothing is broken during the implementation stage. The self-correction features allow to correct some of the problems automatically and provide information to developers about some corrections that they must implement. When changes are made in the applications data models, APIs and architecture, the OutSystems platform automatically updated all existing dependencies. Consequently, the number of bugs introduced is generally lower when compared to traditional development technologies. On the other hand, it is important to consider that low-code platforms have enabled many professionals from other areas to join development teams and that eventually their lack of specific skills may translate into an increase in bugs in the software developed. Despite the support provided by the platform, there is no guarantee that errors will not exist and that the software under development will be bug free. Thus, throughout the life cycle of an OutSystems application, there are several moments when testing activities must be performed. Component Test (Unit Test) are used to verify the behavior of code units (components). In OutSystems, code units are Actions, elements which implement the business logic. It is important to refer that code units are not often easily accessible to be tested. These tests are delivered by the developers as part of the activities developed in the sprint performed in the development environment (DEV) as well as in the continuous integration environment (CI). Usually, they are automated tests.

88

J. Salgueiro et al.

Integration/API tests are tests to verify the integration with external systems. These tests are very important since low-code platforms have simple ways to use external API and many OutSystems applications use external APIs. They must be performed in the DEV environment by the developers or in the quality Assurance (QA) environment in case there are integrations that are not reliably available in the development environment. These tests are often a good target for automation. System Tests or End-to-End (E2E) are tests that are performed considering the perspective of the end user or the system. These tests are usually run through a web or mobile interface. Not all E2E tests are UI (user-interface) tests. This type of tests can be automated by developers, if they are not UI tests, or they can be automated tests performed by the quality team in the case of UI tests. These tests must be performed in the quality QA environment. Acceptance Tests are done by the company / client before approval of a new set of changes or before the application goes into production. Most of the cases addressed are critical resources for the business, and normally, they are performed manually in the QA environment but they can also be performed, manually or automated, in the DEV environment by the developers before the features are passed on to the next environment. Regression Tests are also carried out to verify if the previously developed functionalities continue to be well executed in newer versions of the applications. As new features are added, the need for automated regression testing also increases. These tests must be performed in the CI environment and in the QA environment. Performance Tests are performed to verify the behavior of the application in terms of performance, namely testing the behavior of the software in relation to the number of users and with large amounts of data. Security Tests test the system security by exploiting common security vulnerabilities at run time. 2.3 Test Automation on Low-Code Development Platforms The specific features of LCDP raise a set of challenges in low-code testing. Some of these challenges are identified in [8] namely: the role of citizen developer and its lowlevel technical knowledge in the testing activities; the importance and consequently the challenges in offering high-level test automation; and leveraging the cloud for executing tests alongside supporting testing of cloud-based applications. The implications that testing activities, mainly test automation, have on the low-code development process is also very important. As described in Sect. 2.2, automation is possible on different kinds of tests such as unit tests, Integration/API tests, and System/E2E tests, etc. If, on the one hand, test automation allows continuous quality assessment, and is essential in Agile and low-code development, in the other hand, it is important to assess the implications that the test automation process will have on the development process.

3 Related Work Test automation is an important topic regardless of the methodologies or technologies used in the software development process. However, the methodologies used are decisive in how the test automation process should be approached and implemented. Specifically,

Best Practices for OutSystems Development

89

the literature relating to software testing in LCDP is very scarce. This may be associated with the fact that these platforms are often associated with agile development, which have implications for the way tests are approached, but it is also due to the fact that, in certain situations, the testing process may derail some of the benefits associated to the low-code development. However, some research work has been done. A study [8] analyze five commercial LCDP (Mendix, Power Apps, Lightning, Temenos Quantum, OutSystems) to identify low-code testing advancements from a business point of view. They analyze the testing facilities embedded in each platform and they identify some challenges when testing low-code software. There are also some works that addressed unit testing in low-code platforms. A mocking solution prototype for the OutSystems LCDP was presented in [9]. The proposed mocking mechanism removes dependencies to components that the developer wants to abstract a test from, as for instance web services or other pieces of logic of an application. In [10] a framework for automated test case generation in the OutSystems platform is proposed. This framework targets the Core Business layer in OutSystems proposed 4-Layer Canvas. Usually, low-code platforms also provide some tools for test automation. OutSystems provide a BDD Framework1 that is an open-source application that provides a set of tools for producing BDD Test Scenarios and can also be used for automated testing. Mendix offers the Mendix Application Test Suite2 that is a suite of tools for embedding testing in the application lifecycle. These tools are built in Mendix, on top of Selenium. Testing in Power Apps can be performed with test studio3 that is developed specifically to support automated end-to-end UI testing of an application designed based on canvas. In fact, test automation enables continuous quality assessment and is essential. However, there are few research papers that address test automation on low-code platforms. Automation is possible on all test levels. It can also be used in functional and nonfunctional test types. As low-code applications use many integrations to other services using APIs, besides system/ End-to-End tests, automated integration/API tests are also essential. Thus, it is important to study the impacts that low-code platforms have on software quality assurance and automation testing activities, and it is important to understand its implications and assess its impacts, about whether and how to automate testing and what tools will be most appropriated.

4 Test Automation in the OutSystems LCDP This section describes some experiments that involve the design of tests for applications developed in OutSystems and which were implemented using different test automation tools. The purpose of the experiments is to assess the extent to which each of the tools allows to implement and execute the tests automatically and to what extent the use of some best practices during the development process have influence on the testing automation process.

1 https://www.outsystems.com/forge/component-overview/1201/bddframework. 2 https://www.mendix.com/evaluation-guide/app-lifecycle/test-automation-qa. 3 https://docs.microsoft.com/en-us/powerapps/maker/canvas-apps/test-studio.

90

J. Salgueiro et al.

4.1 Test Automation Tools for OutSystems The experiments described in this section include testing the implementation and execution with two test automation tools: BDD Framework and Ghost Inpector. These are the two test automation tools used in this study although others could have been included such as Selenium, Katalon, JMeter, and others. The BDD Framework was selected because it is an open-source framework designed for test automation in the OutSystems LCDP. The BDD Framework [11] is a testing framework that provides a set of tools for easily creating BDD/TDD tests for an OutSystems application. It is a component that facilitates test-automation where tests are specified using the Gherkin syntax. This framework can be used for Traditional Web and Service apps, the server component of Reactive Web or Mobile apps, and REST and SOAP APIs. Ghost Inspector4 was selected because it is integrated with the framework to support automated Web UI testing and monitoring in OutSystems. 4.2 Experiments Next, some examples are described to illustrate the impacts that some best practices in OutSystems development can have as facilitators in the test automation process. Integration/API Tests The public API that we will target in these tests is restcountries.eu. It is a simple RESTful API, where it is possible to obtain data on all countries in the world. We will use the next scenario: Scenario (Test Case 1): Obtaining the country’s capital. Given: That there is a country called Portugal When: I request data about Portugal Then: The city of Lisbon must be given as its capital The BDD Framework has some web blocks that are used to specify the test steps. In Fig. 1 is possible to see the test scenario using the BDD Framework.

Fig. 1. Test case 1 using BDD Framework.

4 https://ghostinspector.com/.

Best Practices for OutSystems Development

91

Each step of the test will have an action with the logic to execute the test. When we are in the scope of a test run and we need to use simulated services to isolate the test application from the remote system (where the API is exposed), we will have to change each module that consumes that specific API. This would be detrimental to maintenance. To avoid this problem, we should isolate the consumption of the API in a wrapper module that exposes the methods of the API through public actions so that other modules can use them. So, in the next images it is possible to see the action that consumes the API (Fig. 2 left) that is public and proper to be used by other modules (Fig. 2 center). In this case, the public action is used in the action associated to the second step of the test. After getting the capital of Portugal we do an assert to validate that the returned capital is the right one (Fig. 2 center right and right).

Fig. 2. Left: consumption of the API in a wrapper module. Center: access the API through a public method. Center Right and Right: test assertion.

System Tests or End-To-End (UI Tests) For these kinds of tests, we can only use Ghost Inspector because we need to interact with the interface. This tool will record the user steps in the browser and will create a script for the test. For this, we will the next scenario (Test case 2): Given: I have an account at “The Wine Club” When: I click on “Login”; And fill in the email and password; And I click “Sign in” Then: The operation must be successful; And I must be redirected to the main page of the application Figure 3 shows the script created by the tool after the recording. Although the test was successfully performed this could have gone wrong, for example, the first step of the test is for the user to click on a link that contains the text “login” but nothing guarantees that there will be no more links that contain the same text on the page. This could mislead and fail the test. Another aspect that is possible to see in the image is that the elements identified have IDs generated automatically by the platform which can lead to the failure of the test and does not facilitate the reading of the test.

92

J. Salgueiro et al.

Fig. 3. Test case 2 steps using Ghost Inspector.

To solve this problem, the developer must correctly identify the elements used in the interaction with the user in order to be able to identify the elements in a simpler and easier way. Figure 4 shows how the developer can give an identifier to the link that will be clicked by the user.

Fig. 4. Personalizing element identifiers.

After all the elements are well identified, the test should be recorded again, and the selectors will be different. The tester can also edit the selectors so the script can be read easily. Figure 5 (Left) shows the script already with the changes and it is possible to see the differences. The test executed before can also be executed with the BDD Framework but cannot be called as UI test because with the BDD it is not possible to access the browser. It is only possible to test the business logic. In this case, with BDD the scenario just needs to have an action that calls the login action that is already developed in the platform and assert if the login was successfully done (see Fig. 5 right).

5 The Influence of Best Practices in Development in Low-Code Software Testing Automation In any programming language, or development platform, applications must be developed to facilitate tests that validate that the software complies with the original requirements. In OutSystems, developing code to be easily testable, and to facilitate the test automation

Best Practices for OutSystems Development

93

Fig. 5. Left: Test case 2 using Ghost Inspector. Right: Test case 2 using BDD Framework.

process, implies the use of some best practices during the software development process. Some of these practices were explored in the previous section, through the test cases presented, and are analyzed and discussed in more detail in this section. In integration/API tests it is important to isolate the API consumption in a specific module that exposes the API methods through public actions. This approach will allow other modules, that need access to the API, to do so through this specific module, avoiding implement and execute tests on every module that is consuming this specific API. Thus, during the test automation process, it will only be necessary to do it in the module that exposes that API. Usually, in System/End-to-End tests automating web UI tests involves simulating the interactions a user would perform, through a browser over the web application, to complete the functionality under test. To be less laborious and successful, it is important that the test automation tools, which are being used, can correctly identify the different web elements found on the web page being tested. To achieve this, it is necessary that the names/IDs of the components are easily identified, which often implies the use of personalized identifiers in place of the identifiers that are usually assigned by the development platforms. On the OutSystems platform it implies that those elements need to be properly identified in Service Studio by the developer. In this case the developer should customize the elements’ names, that will be the ID of the html element, to ensure that all elements that have some degree of user interaction have an identifier that would be uniquely identified during the test automation process. However, if on the one hand customizing element identifiers while developing facilitates the test automation process, on the other it requires time and more resources and can be a complex task for developers without specialized skills. In fact, the implementation of best practices during the development process can have a significant influence on the test automation process. However, it is important to consider whether the workload associated with the test automation process and the implications it

94

J. Salgueiro et al.

implies for the adoption of certain practices in development can undermine the benefits of low-code platforms without a low-code UI automation tool. Additionally, it must be considered that running a test case takes time on either tool. Since to run the tests with the BDD framework it is necessary to have knowledge of the OutSystems and the tests must be developed by the developers. For this, each user story must have time allocated to develop the tests. On the other hand, with Ghost Inspector tests can be performed by users without knowledge of OutSystems but it does not invalidate technical knowledge since it may be necessary to edit the JavaScript selectors. BDD Framework has the advantage of being developed in the same language as the platform but does not allow UI tests and the tests must be carried out at the level of business logic. An automated web UI testing tool, such as the Ghost Inspector, is easy to use and allows the recording of test crypts without the need to know code details. However, in order for it to be effectively used and allow the tests to be executed in an automated way, mainly the regression tests, it may be necessary that the developer has to make changes to the code developed by the platform.

6 Conclusions and Future Work This work approached the test automation process on the OutSystems LCDP. A set of test cases are described to illustrate the impacts that best practices in the development of OutSystems applications can have as facilitators in the test automation process and in the feasibility of using different testing tools. The test cases are implemented and executed using two test automation tools: BDD Framework and Ghost Inspector. Examples showed that the use of some best practices during the development in OutSystems can have a significant impact on the test automation process. As future work we intend to analyze in more detail which tool is best suited to the test automation process in OutSystems.

References 1. OutSystems: State of Application Development Report (2019) 2. Marqual IT Solutions Pvt. Ltd (KBV Research): Global Low-Code Development Platform Market By Component By Application By Deployment Type By End User By Region, Industry Analysis and Forecast, 2020–2026, Report (2020) 3. OutSystems: Low-Code Development Platforms. https://www.outsystems.com/low-code-pla tforms/. Accessed 30 Sept 2020 4. Boulton, C.: What is low-code development? A Lego-like approach to building software. https://intellyx.com/2018/03/27/what-is-low-code-development-a-lego-like-approachto-building-software/. Accessed 07 Aug 2020 5. Idle, J.: Low-Code rapid application development - So, what‘s it all about? (2016) 6. Marie Walsh, Cornanguer, B., Forgács, I., Kakkonen, K., Coleman, G., Sabak, J., Black, R.: Agile Testing Foundations. An ISTQB Foundation Level Agile Tester Guide (2017) 7. International Qualification Board for Business Analysis: Certified Tester Specialist Syllabus Foundation Level Acceptance Testing (2019)

Best Practices for OutSystems Development

95

8. Khorram, F., Mottu, J.-M., Sunyé, G.: Challenges & opportunities in low-code testing. In: Proceedings of the 23rd ACM/IEEE International Conference on Model Driven Engineering Languages and Systems: Companion Proceedings, New York, NY, USA. Association for Computing Machinery (2020). https://doi.org/10.1145/3417990.3420204 9. Jacinto, A., Lourenço, M., Ferreira, C.: Test mocks for low-code applications built with OutSystems. In: Proceedings of the 23rd ACM/IEEE International Conference on Model Driven Engineering Languages and Systems: Companion Proceedings, New York, NY, USA. Association for Computing Machinery (2020). https://doi.org/10.1145/3417990.3420209 10. Gomes, A.: An automated unit testing framework for the OutSystems Platform (2018) 11. Proença, J.: BDD Framework: overview. https://www.outsystems.com/forge/component-ove rview/1201/BDDFramework. Accessed 13 Nov 2020

Solving the Multi-objective 2-Dimensional Vector Packing Problem Using -constraint Method Nadia Dahmani1,2(B) , Saoussen Krichen2 , El-Ghazali Talbi3 , and Sanaa Kaddoura1 1

3

College of Technological Innovation, Zayed University, Abu Dhabi, UAE [email protected] 2 Institut Sup´erieur de Gestion de Tunis, LARODEC, University of Tunis, Tunis, Tunisia Universit´e de Lille 1, LIFL CNRS UMR 8022, INRIA Lille-Nord Europe, Villeneuve-d’Ascq, France Abstract. In this paper, an exact method is designed to solve the multiobjective 2-dimensional vector packing problem. The algorithm is an adapted version of an efficient -constraint method which proves its efficiency in solving a large variety of multi-objective optimization problems. This method is based on a clever decomposition of the initial problem into sub-problems which are iteratively solved through mathematical programming. To accelerate the search process, we propose a new integer programming model for solving the multi-objective 2-dimensional vector packing problem based on the compact model for the bin packing problem with fragile objects. Instead of scanning all possible solutions, we consider the solutions while solving a Subset-Sum Problem. Hence, nonuseful subproblems are avoided and thus the search space is reduced. An experimental study is performed based instances from the literature. A comparison between the exact method and a grounded metaheuristic which provides good results in solving the multi-objective 2-dimensional vector packing problem. Keywords: 2-dimensional vector packing problem · Multi-objective combinatorial optimization · -constraint method · Metaheuristics

1

Introduction

The multi-objective 2-dimensional vector packing problem (MO-BPP) deals with packing a set of items each having two sizes in two disctinct dimensions (e.g. a weight and a height) into an unlimited number of identical bins with capacity sizes. Three objectives are concurrently optimized. The first objective is the minimization of the number of used bins. The second one is the minimization of the maximum height of a bin. The third objective consists in balancing the load overall the bins by minimizing the difference between the maximum height and the minimum height of a bin. c The Author(s), under exclusive license to Springer Nature Switzerland AG 2021  ´ Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 96–104, 2021. A. https://doi.org/10.1007/978-3-030-72654-6_10

Solving MO-BPP Using -constraint Method

97

Apart from its theoretical importance, the MO-BPP has a practical relevance since various real-world applications can be modeled as multi-objective packing problems (see among others [3,10,13,16]). As a generalization of the conventional one-dimensional bin packing problem, MO-BPP is N P-hard in the strong sense [8]. Dahmani et al. [4] investigated this problem and suggested self-adaptive metaheuristics so as to provide a good approximation set. Yet, the need for benchmarking, which is often hard without the knowledge of the true Pareto set, guide us to explore exact algorithms in order to find all the Pareto-optimal solutions. Despite existing various exact methods for multi-objective problems encompassing many specialized algorithms solving specific problems, those methods were often designed to tackle bi-objective problems (see [5,7,14]). Indeed, only few of them handle multi-objective problems which are extended from the well-known constraints method [9]. This method mainly transforms the multi-objective problem into a suite of parameterized single-objective problems in a way that the optimum of each single-objective problem refers to a Pareto-optimal solution. The major asset of the -constraints method concerns the achievement of efficient points in a non-convex Pareto curve. Indeed, compared with the weightingsum method that generates only efficient extreme solutions, the -constraints method alters the original feasible region and is able to produce non-extreme efficient solutions. Furthermore, non-supported efficient solutions can not be generated with the weighting method, while the -constraints method does not suffer from this issue. Many improvements for the -constraint method have been proposed in the literature in order to enhance its performance or adapt it to specific multi-objective problems. The main contribution of this paper consists in proposing an efficient epsilonconstraint method to generate all non-dominated solutions composing the Pareto front for MO-BPP. This method is an adapted version of the augmented constraint method (AUGMECON2) [12] which proves its efficiency in solving several multi-objective optimization problems. The algorithm proceeds by decomposing the initial problem into subproblems which are iteratively solved through mathematical programming. At each iteration, it exploits the information from the slack or surplus variables so as to avoid many redundant iterations. Hence, the computation time is reduced. In order to accelerate the search process, we introduced a new mathematical formulation of the MO-BPP based on the compact model for the Bin Packing Problem with Fragile Objects [2]. Instead of scanning all possible solutions, we consider the solutions while solving a Subset-Sum Problem. Hence, non-useful subproblems are avoided and thus the search space is reduced. The performance of the proposed method was assessed using various benchmarks inspired from the literature [1]. Differents quality measures have been used to compare the results of the exact method AUGMECON2 with those of the multi-objective metaheuristic NSGA-II which proves its efficiency in solving the MO-BPP [4]. The reminder of the paper is organized as follows. Section 2 is devoted to the MO-BPP by providing a formal definition and a mathematical formulation of the addressed problem. Section 3 deals with the proposed solution approach. In

98

N. Dahmani et al.

Sect. 4, we report our experiments. A conclusion and perspectives are drawn in the last section.

2

The Multi-Objective 2-Dimensional Vector Packing Problem

In MO-BPP, we are giving a set of N items j, each having a weight cj and a height hj (j = {1, . . . , n}) and an unlimited number of identical bins with two capacity sizes C and H. In what follows, we state a new mathematical formulation for the MO-BPP based on the compact model proposed in [2] for the Bin Packing Problem with Fragile Objects. The model (2)–(11) uses the following decision variables  • xji =

1 if item j is assigned to the bin having item i as item with smallest index, 0 otherwise

 • xi =

1 if item i is the item with smallest index, 0 otherwise

• H: Integer variable that expresses the maximum height loaded into one bin. • E: Integer variable that expresses the minimum height loaded into one bin. • D: Integer variable that expresses the difference between H and E (i.e., H −E).

min f1 (s) =

N 

xi

(1)

j=1

f2 (s) = H f3 (s) = D s.t xi +

i−1 

xij = 1, ∀i ∈ {1, . . . , N },

(2) (3) (4)

j=1 N 

wj xji ≤ (C − wi )xi , ∀i ∈ {1, . . . , N },

(5)

hj xji ≤ H − hi xi , ∀i ∈ {1, . . . , N },

(6)

hj xji + hi xi + (1 − xi ) × θ ≥ E, ∀i ∈ {1, . . . , N },

(7)

j=i+1 N  j=i+1 N  j=i+1

xji ≤ xi , ∀i ∈ {1, . . . , N }, ∀j ∈ {i + 1, . . . , N },

(8)

xji ∈ {0, 1}, ∀i ∈ {1, . . . , N }, ∀j ∈ {i + 1, . . . , N }, xi ∈ {0, 1}, ∀i ∈ {1, . . . , N }

(9) (10)

{H, E, D} ∈ N. s = (xji , xi ), ∀i ∈ {1, 2, . . . , N }, j ∈ {i + 1, . . . , N }

(11) (12)

Solving MO-BPP Using -constraint Method

99

Objective Functions • The first objective (1) seeks to minimize the number of used bins. • To get the second objective (2), the height capacity constraint is relaxed and the maximum height of a bin H is minimized. • The third objective (3) tries to minimize the maximum height H of a bin. System Constraints • Constraints (4) impose that either an item is the smallest item in its bin, either it is assigned to a bin containing an item with smaller index. • Constraints (5) require that the sum of the weights of the items packed in a bin does not exceed the smallest index in the bin. • Constraints (6) require that the sum of the heights of the items packed in a bin does not exceed the smallest index in the bin. • Constraints (7) use θ as a big number so that xi can function as a switch to turn constraints (7) on and off. • Constraints (8) are used to tighten the model linear relaxation.

3

The Solution Approach

In this section, a adaptive version of the AUGMECON2 method to the 2dimensional multi-objective vector packing problem is presented. For any  ∈ R2 , the resultant problem (13)–(23) can be stated as follows. z = min

N 

xi + β × (s2 /r2 + 10−1 × s3 /r3 )

(13)

j=1

s.t xi +

i−1 

xij = 1, ∀i ∈ {1, . . . , N },

(14)

j=1 N 

wj xji ≤ (C − wi )xi , ∀i ∈ {1, . . . , N },

(15)

hj xji + hi xi ≤ H, ∀i ∈ {1, . . . , N },

(16)

hj xji + hi xi + (1 − xi ) × θ ≥ E, ∀i ∈ {1, . . . , N },

(17)

j=i+1 N  j=i+1 N  j=i+1

xji ≤ xi , ∀i ∈ {1, . . . , N }, ∀j ∈ {i + 1, . . . , N }, H ≤ 2 , ∀i ∈ {1, . . . , N },

(18) (19)

H − E ≤ 3 , xji ∈ {0, 1}, ∀i ∈ {1, . . . , N }, ∀j ∈ {i + 1, . . . , N },

(20) (21)

xi ∈ {0, 1}, ∀i ∈ {1, . . . , N } x ∈ X.

(22) (23)

100

N. Dahmani et al.

where, ri , i = 2, 3 represent the range of objective i which is computed from the payoff table of the original MO-BPP using the Ideal and Nadir values of H + , H − , D+ , D− . Note that x ∈ X represents the feasible region of the original MO-BPP (i.e., Eqs. (4)–(10) ). β is a small positive value usually fixed between 10−3 and 10−6 . In real-world problem with k objectives, like ours which has 3 conflicting objectives, the model (1)–(11) results in k − 1 -constraints. Several aspects are easy in the bi-objective case (e.g. computation on the Nadir point, decomposition of the search space) but require to be carefully studied in the multi-objective case. 3.1

Ideal and Nadir Points Computation

To find the entire efficient solutions set, the search space must be delimited using the Ideal and the Nadir points. These points indicate extreme values for each objective. The computation of such values is easy for the bi-objective case (i.e. using the two extreme points). Although the Ideal point is still relatively easy to find in the multi-objective framework, it is not the case for the Nadir point. To get the Ideal point (I) with K objectives, the K following single objective problems must be minimized. ∀k ∈ K, fk (I) = min fk (x). x∈K

Similarly, the Nadir point (N ) can be obtained by maximizing each objective function over X . The Nadir point (N ) cannot be computed easily for problems with K objectives. To get this point, interesting properties of the set OptK−1 are given in [15]. Theorem 1. Given a multi-objective optimization problem, the set OptK−1 has the following properties: (1) OptK−1 contains a set of K-Pareto solutions such that for every component of the objective vector, the maximal value of any K-Pareto solution is found within this set. (2) The Nadir point (N ) can be computed by: ∀k ∈ K, fk (N ) =

max

x∈OptK−1

fk (x).

Based on this theorem, an efficient algorithm for computing the Nadir point is stated in [6]. The algorithm is based on a recursive computation of the sets OptK−k , ∀k < K. Once the Ideal and the Nadir points have been computed, the AUGMECON2 method can be performed for each grid point in the objective function ranges. 3.2

Reducing the Search Space Size

Based on the special structure of our problem, we notice that not all values of H are possible. Some do not correspond with the sum of heights of any subset

Solving MO-BPP Using -constraint Method

101

of items. Therefore, to ensure that a value H is useful, we solve a Subset-Sum Problem (SSP): Given a set of N positive integers and a value C, the objective of this problem is to find the subset of integers whose sum is the closest to C without exceeding C. If this problem has no solution for value H, then this value is skipped. Instead of scanning linearly all possibles values of H ∈ [H − , H + ], we select the values of H that for an SSP(H) a solution exists. Hence, the computation of many non-useful subproblems is avoided and thus the search space is reduced.

4

Experiments

We empirically investigate the performance of our -constraint method by comparing its the results with NSGA-II. The choice of NSGA-II was approved by the results presented in [4] where the algorithm exceeds other competitive multiobjective algorithms in solving MO-BPP. 4.1

Software-Hardware Implementation

We coded the proposed solution approach using CPLEX 12.0. Hence, the resulting integer linear program (17)-(26) was exactly solved using a branch-andbound algorithm. NSGA-II was implemented in C++ using the ParadisEO framework [11]. All the experiments have been conducted on a an Intel Core 2 Duo 6600 (2.40 GHz) computer, and 2 GB RAM running Linux. The stopping criterion for NSGA-II is related to the computational time. Based on [4], it was fixed to 60 s for a single simulation run per instance. 4.2

Computational Results

In this section, a comparative study for quality measures is carried out for both AUGMECON2 and NSGA-II algorithms in solving the MO-BPP using different test instances from Caprara and Toth [1]. While the exact method AUGMECON2 requires only one run of the whole algorithm for each instance, NSGA-II results are averaged over 20 independent runs. The grid points for AUGMECON2 considers the range of the second objective function which is increased by one at each iteration. Table 1 summarizes the computational results and provides the following information: Let F ∗ denotes the Pareto optimal set generated by the AUGMECON2 method and P the approximation set generated by NSGA-II. • |F ∗ |: the number of efficient solutions in the exact Pareto front; • N b pbs: the number of integer programming problems solved for each instance; • T ime: the total computational time; • |P |: the number of non-dominated solution in the approximation set; • Cov: the average coverage metric value [17]. This quality measure reports the percentage of non-dominated solutions in P covered by those in F ∗ ;

102

N. Dahmani et al.

• Eps: the average unary additive -indicator value [17]. This convergence based metric gives the minimum value by which P has to be translated in the objective space to weakly dominate F ∗ ; • Hyp: the average unary hyper-volume metric value [17]. This quality indicator informs about both convergence and diversity. It computes the portion of the objective space that is weakly dominated by F ∗ and not by P .

Table 1. Computational results of the AUGMECON2 method

Instance

AUGMECON2 |F ∗ | Nb pbs T ime

NSGA-II |P | Cov

Eps

Hyp

CL 01 25 1

5

357843

21 h 34 min

8

35,5%

CL 01 25 2

4

389321

10 h 51 min

8

61,81% 10,45 11,07

26,87 17,98

CL 02 25 1

3

406393

41 h 27 min

6

CL 02 25 2

6

467454

11 h 37 min 10

72,72% 9,15

CL 03 25 1

12

190534

20 h 2 min

9

16,66% 32,19 38,89

CL 03 25 2

14

206108

16 h 34 min

9

CL 04 25 1

13

190534

15 h 23 min 10

39,56% 32,19 40,75

CL 04 25 2

15

206108

13 h 67 min 13

42,87% 15,76 29,93

CL 06 25 1

16

7718

41 h 22 min

8

37,5%

17,51 16,21 20,81 26,01

36,37% 30,68 25,97

40%

9,04

15,76 20,06

CL 06 25 2

9

7691

23 h 40 min

8

40%

CL 10 25 1

19

9679

8 h 12 min

6

21,05% 25,95 14,67

CL 10 25 2

17

9632

7 h 33 min

13

52,94% 16,74 17,84

CL 10 25 3

13

9406

11 h 53 min 13

53,84% 12,9

CL 10 25 4

15

9416

33 h 6 min

15

60%

CL 10 25 5

19

9580

5 h 33 min

13

47,36% 18,22 13,64

CL 10 25 6

17

9658

8 h 47 min

14

15,78% 31,2

CL 10 25 7

14

9403

13 h 27 min 12

50%

CL 10 25 8

16

9406

16 h 50 min 13

56,25% 13,15 11,85

CL 10 25 9

16

9688

6 h 30 min

12

35,29% 28,53 26,97

CL 10 25 10 17

9661

5 h 40 min

13

47,05% 16,46 15,49

14,08

11,01 10,28 28,37

15,38 12,51

Regarding the AUGMECON2 algorithm, it can be seen from the results that the size of the Pareto set, and therefore the total running time increases very quickly with the difficulty of the corresponding instances class. This makes very difficult to solve larger instances of the MO-BPP to optimality. Obviously, the running time of the branch-and-bound algorithm increases considerably in later iterations, when the objective space becomes more dense and the constraint bounds become very tight. Nevertheless, it is noteworthy that to the best of our knowledge no exact Pareto fronts have been computed so far

Solving MO-BPP Using -constraint Method

103

Fig. 1. Pareto fronts comparison

for the multi-objective vector packing problem, probably due to the lack of an appropriate generating method besides complete enumeration. The results obtained with NSGA-II show that it generates relatively good approximations of the exact Pareto fronts for a very short time (60 s) compared with AUGMECON2. However, it has difficulty in approximating the entire Pareto front even with small instances. NSGA-II achieves an average coverage metric (Cov) between 16,66% and 72,72%. A lower average value for both Eps and Hyp indicators signifies a “better” approximation set. It can be noted from the results that the difference can be large for both metrics with regard to the difficulty of the test instance. The values vary from 9,15 and 9,05 for instance CL 02 25 2 and 32,9 and 38,89 for instance CL 03 25 1, according to Eps and Hyp respectively. Furthermore, there exist specific runs where whole portions of the Pareto front where not discovered (see Fig. 1).

5

Conclusion

In this paper, a solution approach based on an adapted version of the efficient -constraint method AUGMECON2 was proposed for solving the MO-BPP. This method aims to generate the whole efficient solutions composing the Pareto front using mathematical programming. In order to accelerate the search process, a new mathematical formulation for our multi-objective problem is stated. An experimental study was accomplished to investigate the performance of the efficient -constraint method based on a comparision of the generated results with the multi-objective metaheuristic NSGA-II which proves its efficiency in solving the MO-BPP and using a well-known benchmark instance. Depite the proposed method was time consuming, the overall results were promising for small size problems. Nevertheless, more research is needed in order to compare AUGMECON2 with other recently available exact algorithms for multi-objective problems (e.g. a two-phase method). It can also be useful to assess the performance of AUGMECO2 with problems with more objective functions and to use specific heuristics to speed-up the exact method.

104

N. Dahmani et al.

References 1. Caprara, A., Toth, P.: Lower bounds and algorithms for the 2-dimensional vector packing problem. Discret. Appl. Math. 111(3), 231–262 (2001) 2. Clautiaux, F., Dell’Amico, M., Iori, M., Khanafer, A.: Lower and upper bounds for the bin packing problem with fragile objects. Discret. Appl. Math. 163(1), 73–86 (2014) 3. Dahmani, N., Clautiaux, F., Krichen, S., Talbi, E.G.: Iterative approaches for solving a multi-objective 2-dimensional vector packing problem. Comput. Ind. Eng. 66(1), 158–170 (2013) 4. Dahmani, N., Clautiaux, F., Krichen, S., Talbi, E.G.: Self-adaptive metaheuristics for solving a multi-objective 2-dimensional vector packing problem. Appl. Soft Comput. 16, 124–136 (2014) 5. Ehrgott, M.: A discussion of scalarization techniques for multiple objective integer programming. Ann. Oper. Res. 147, 343–360 (2006) 6. Ehrgott, M., Tenfelde-Podehl, D.: Computation of ideal and Nadir values and implications for their use in MCDM methods. Eur. J. Oper. Res. 151, 119–139 (2003) 7. Ehrgott, M., Gandibleux, X., Przybylski, A.: International series in operations research and management science. In: Exact methods for Multi-objective Combinatorial Optimisation, pp. 817–850. Springer (2016) 8. Garey, M.R., Johnson, D.S.: Computers and intractability: a guide to the theory of NP-completeness (Series of Books in the Mathematical Sciences). Freeman, W. H (1979) 9. Haimes, Y., Lasdon, L., Wismer, D.: On a bicriterion formulation of the problems of integrated system identification and system optimization. 1(3), 296–297 (1971) 10. Levin, M.S.: Bin packing problems (promising models and examples). J. Commun. Technol. Electron. 63(6), 655–666 (2018) 11. Liefooghe, A., Basseur, M., Jourdan, L., Talbi, E.G.: ParadisEO-MOEO: a framework for evolutionary multi-objective optimization. In: Obayashi, S., Poloni, C., Deb, K. (eds.) Evolutionary Multi-Criterion Optimization, Fourth International Conference (EMO 2007), vol. 4403 of LNCS., Matsushima, Japan, pp. 386–400. Springer (2007) 12. Mavrotas, G., Florios, K.: An improved version of the augmented -constraint method (augmecon2) for finding the exact pareto set in multi-objective integer programming problems. Appl. Math. Comput. 219(18), 9652–9669 (2013) 13. Naderi, B., Yazdani, M.: A real multi-objective bin packing problem: a case study of an engine assembly line. Arab. J. Sci. Eng. 39(6), 5271–5277 (2014) 14. Turgut, O., Dalkiran, E., Murat, A.E.: An exact parallel objective space decomposition algorithm for solving multi-objective integer programming problems. J. Global Optim. 75, 35–62 (2019) 15. Reeves, G.R., Reid, R.C.: Minimum values over the efficient set in multiple objective decision making. Eur. J. Oper. Res. 36(3), 334–338 (1988) 16. Spencer, K.Y., Tsvetkov, P.V., Jarrell, J.J.: A greedy memetic algorithm for a multiobjective dynamic bin packing problem for storing cooling objects. J. Heuristics 25(1), 1–45 (2019) 17. Zitzler, E., Thiele, L., Laumanns, M., Fonseca, C.M., da Fonseca, V.G.: Performance assessment of multiobjective optimizers: An analysis and review. IEEE Trans. Evol. Comput. 7, 117–132 (2002)

A Systematic Review of Risk Management in System Dynamics Project and Techniques Eliane Cunha Marques(B) , Marcos Vinicius L. Castro(B) , Hilario Luiz Babireski Junior(B) , and Ricardo Matos Chaim(B) Universidade de Brasília, Brasília, DF 70910-900, Brasil {mvlcastro,ricardoc}@unb.br

Abstract. The objective is to analyze how the literature related to risk management in projects using System Dynamics deals with the causal relationships between risk factors in order to assist the project manager. It concludes with considerations about the use of System Dynamics in the effective management of risks in projects. Keyword: Project management · Risk management · Systems dynamics

1 Introduction Systems Dynamics models have been widely used in projects and have proven their value in contributing to significantly better design performance [1]. System Dynamics combined with Risk Management is applicable to complex systems, creating a management tool that enables decision making, minimizing uncertainties and maximizing opportunities [2]. The globalization and the interdependence of economic, social, environmental and cultural systems, packed in the speed of technological development, create an environment of risks and uncertainties for the manager to make decisions inherent to his job. These challenges are amplified by the complexity involving a universe of variables that affect the impact of the result of a decision. Reducing uncertainty is expensive, because making the right decisions requires information [6]. A project is a single, complex and connected sequence of activities to achieve an objective or purpose, which must be completed within a period, within a defined budget and according to the technical specification to meet the requirements of the clients [16]. Project Risk Management includes the processes of planning, identification, analysis, response planning and risk control [16]. The risk analysis includes the management of the probability and the impact of positive and negative events. The purpose of this article is to analyze how the literature on risk management in projects using System Dynamics deals with the causal relationships between risk factors in order to assist the project manager. Therefore, the systematic review of the literature sought answers to the following questions: What risk factors need immediate action to reduce/mitigate risk in project management? Is System Dynamics a method that enables the Project Risk Management process? Does System Dynamics provide support for the © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 105–118, 2021. https://doi.org/10.1007/978-3-030-72654-6_11

106

E. C. Marques et al.

decision process? The systematic review is the scientific investigation that brings together relevant studies on a question formulated, using the database of literature that deals with that question as a source and methods of identification, selection and systematic analysis, in order to carry out a critical review and comprehensive literature [3, 4]. The theoretical section on Risk Management and Systems Dynamics applied to projects is presented in the following section. In addition, the sources of risk in the projects are discussed. Next, the methods for developing a systematic literature review are presented. Subsequently, the selected articles are reviewed and discussed. Finally, the conclusions, limitations of the article and recommendations for future research are presented.

2 Theoretical Framework 2.1 Risk Management Applied to Project Project Risk Management aims to identify and prioritize risks before they occur, in addition to providing action or guidance information to project managers. This orientation requires the consideration of events that may or may not occur and, therefore, are described in terms of probability or probability of occurrence, in addition to other dimensions, such as their impact on objectives. Risks can affect the project’s life cycle, costs, financial performance and strategic objectives [21]. In addition, stakeholders are increasingly demanding Project Risk Management to protect themselves against financial or legal consequences [19]. Therefore, Risk Management is crucial and indispensable for the success of the project, as it can avoid possible problems [20]. Kerzner and Saladis [22] show that the project’s risk is defined as a measure of the probability and consequence of the project’s goal not being achieved. Wang and Chou [23] defined “risk efficiency” as the minimum level of risk for a specific level of expected performance, since risk in projects cannot be eliminated. Risk Management reduces risks by identifying events and developing response plans to monitor project execution. Project Risk Management Practices - ® PMBOK [7] state: All projects are uncertain. Uncertainty is inevitable, since projects are unique and temporary ventures based on assumptions and constraints, delivering results to various stakeholders with different requirements. The Project Management Knowledge Guide [16] is the most popular reference for managing project risk. Although PMBOK is effective in identifying resources, tools and techniques and results, each project needs an appropriate application to manage risks [16]. 2.2 System Dynamics Applied to Projects System Dynamics is a particular application of diagrams that are used to identify risks within a project situation. The System Dynamics model represents entities and information within a project, and the analysis of the model can reveal feed-back and feed-forward loops that lead to uncertainty or instability. In addition, the results of a Systems Dynamics analysis can show the impact of risk events on the overall results of the project, such as cost or schedule results. Analysis of changes in the model or assumptions can indicate the sensitivity of the system to specify events, some of which may be risky. [7] System Dynamics techniques can be applied to understand and influence the way in which

A Systematic Review of Risk Management

107

the elements of a system vary over time. These techniques use concepts from the control field with feedback (feedback loops) to organize the available information about a system, creating models that can be simulated on a computer [8]. Given the dynamic nature of design problems, it is not surprising that System Dynamics has been applied to understanding and improving the behavior of complex projects [9]. System Dynamics focuses more on the importance of feedback loops compared to cause and effect analysis, in which dependent and independent variables are explicitly defined [10, 11]. It helps to understand the dynamic relationship between the variables that use the feedback process and examine the behavior of the entire system through simulation of modeling and analysis [9, 12, 13]. Information Technology (IT) project management is a type of project management where IT projects are planned, executed, monitored and controlled by a project team [14]. In comparison with other types of projects (for example, construction and manufacturing projects), IT projects are usually short-term and with greater uncertainty, they also include cloud computing projects, which are de-challenging. Because of its dynamic nature [15]. 2.3 Project Risk Management Using System Dynamics Many different types of models have been developed to improve project management. These models include some of the system features and characters covered by System Dynamics. For example, basic design models, such as the critical path method, explicitly model causally linked activities and development phases, and cost control models use predicted performance gaps (for example, budget deficits) for all funds. More advanced models, such as the computational models developed by Levitt et al. [26] and others, are very similar to System Dynamics, as they include linked development activities, as well as feedback. The literature on dynamic system design models varies widely in the level of details provided, especially in the descriptions of the model structure, from the complete dissemination of the model to almost none. Some authors focus on the structure of the model, while others focus on its use and describe the structure of the model only in general terms [27–29]. Assessments are necessarily limited when model equations or detailed structure information are not available. However, the review reveals a positive relationship between the access provided by the authors to model details and the subsequent use of these models by other researchers and professionals. The structures that System Dynamics has used to model projects can be described in four groups based on the central concept that they integrate into project models. The category provides a meta-structure of the project model and relates these structures to the Systems Dynamics methodology. The four structure groups of the model are [5]: Project resources, rework cycle, Project control and Undulating and indirect effects. 2.4 Method, Model, Methodology in Project Risk Management We find techniques, methods, models, tools and methodologies influential in Project Risk Management. System Dynamics is a modeling technique used to understand and study the dynamic and complex behavior of systems using numerical data and a graphical expression of the results [11, 43–46]. Sys-tem Dynamics, is also described as an interdisciplinary method to improve learning in complex systems [9]. A model is a

108

E. C. Marques et al.

simplified representation or abstraction of reality that describes, reflects or replicates a real event [39]. The quantitative model is based on a set of variables that vary over a specific domain, while the quantitative and causal relationship were defined between these variables [41]. The models are classified into normative, descriptive and prescriptive. Normative models are used by philosophers, economists, scientists to explore how they should make the decision [37] under rational behavior [40]. Descriptive models are used by ficologists to explain how people make decisions, what theoretical basis has been influenced in studies on the limited capacity of peoples. Prescriptive models are an interaction of the two previous models, including values, preferences, beliefs and judgments of peoples that are consistent in a normative theory [37]. Many project risk management techniques are long overdue [36]. However, these techniques do not integrate all risk actors and all design processes. When analyzing the methods and models available in the literature, the proposal by Marcelino-Sádaba et al. [25] integrate all design and use activities as a basis for the PMI life stages, but also from some procedures [42]. Project management and risk management methodologies cannot be standardized for all types of projects, but must be adapted to the nature of the objectives and uncertainties of each project [24]. In addition, some techniques and tools are used to assess risks and their level of application. For example, Failure Mode and Effect Analysis (FMEA) has a strong application for risk identification, analysis of consequences, probability, level and risk assessment. Markov analysis can only be used to identify risks and analyze risk consequences. Generic approaches to risk management standards also present techniques and methods used to identify and analyze risks, and can be applied in a general context of organizations [18].

3 Literature Review Methodology and Procedures In order to obtain the answers to a problem when there is not enough information, or even when the available information is presented in a disorderly manner and without connection to the problem, research is required [30]. Still on it, it constitutes a set of actions formalized by scientific rigor and treatment in order to lead to reality or to the discovery of partial truths [31]. When the research constitutes a search in printed or electronic bibliographic sources, it consists of a bibliographic search [34]. Due to the characteristics of the activities employed, which consist of collecting bibliographic material for documentary purposes, the nature of the research is considered to be basic. Since the presentation of the data takes place according to verbal descriptions, and the focus is given by the researcher in the interpretation of the data [30], due to the use of a literature review based on scientific articles file, the qualitative approach, based on the interpretation of the explored bibliographic sources. As for its purpose, to make it possible to increase the researcher’s knowledge about the phenomenon to be investigated [34], it is an exploratory type of research. Data collection was performed through searches in the Scopus database, through consultation of articles related to the research topic. The English search terms “risk management”, “project management” and “system dynamics” were used. As the subject area, the option “Business” was selected, and as a time limitation, articles available from 2015 were consulted to ensure the selection of the most recent works. As an impact reference, the articles were classified by number of citations, from the highest to the lowest, in all searches performed (Table 1).

A Systematic Review of Risk Management

109

Table 1. Seach criteria used Search Terms

Options

Publication date: From de 2015 Search Fields:

Titles, Abstract and Keywords

Database used:

Scopus

Search Terms:

“risk management” “project management” “system dynamics”

Search Area

Business

Source: from the authors

Searches on the search platform were classified by the number of citations, to use the material with the greatest impact. The selection was made based on reading the abstracts (Table 2). Table 2. Search results (Scopus) Keywords Searched “risk management” AND “system dynamics”

Documents 96

“project management” AND “system dynamics”

886

“risk management” AND “project management”

386

“risk management” AND “project management” AND “system dynamics”

9

Source: from the authors

Based on the results of the research, articles were selected to prepare bibliographic surveys.

4 Literature Review Analysis and Findings After analyzing the 1,677 articles, it was possible to consider that: (a) most of the tools used in the context of Risk Management are not applicable to the entire Risk Management process [38]; (b) the relevant methods for identifying, analyzing, assessing and dealing with risks, such as brainstorming, are not structured; they deal only with qualitative information and are limited by the users’ experience [39]; (c) risk is generally addressed regardless of the project and its environment and (d) 77 articles identified 149 high-severity risk factors that need immediate risk reduction / mitigation action by the project management team, as classified in Table 3. Project Risk Management is carried out independently and is not integrated with all project activities. It is noted that some methodologies for integrated project risk management have been developed. They are usually based on a temporal representation of the project. Thus, the project

Risk

Failure to integrate project management; lack of support for infrastructure; Decision making; Management support; Supervision of activities; Non-existent or bad planning; Lack of an effective project management methodology; poor or nonexistent control; Adoption of a new method / technology; Use of a new methodology; System with integration and interface with other systems; Failure in project management; No control; Realistic project configuration; Failure to manage systems development; Delivery failure; Lack of componentization; Technical novelty in infrastructure; Project management methodology

Hanges per customer; design mudance due to technology; deviations from scope; Nature of the project; Change in scope/goals; Poorly understood and/or defined requirements; Unclear or mistaken scope / goals; Volatility in requirements (lack of static requirements); Misunderstanding of requirements; Gold Plating; Constant changes in technical requirements; Design / project; Inadequate change management

Deviation of time and overload of schedules; delay in project approvals and licenses; any delays that have occurred; Poorly estimated deadlines and schedules for tasks; Artificial deadlines; Delays in schedules; change in legislation; damage to previously made elements and markings; Climate; Project location; Site conditions; Complex tasks; Change in the characteristics of the activities;

Knowledge areas

Project Integration Management

Project Scope Management

Schedule Management

Selladurai Pitchaimuthu, Jitesh J. Thakkar and P.R.C. Gopal (2019); Xu, Y., Chan, A.P., Xia, B., Qian,Q.K., Liu, Y. and Peng, Y. (2015); Boehm (1991); Nakashima and Carvalho (2004); Schmidt et al. (2001); Wallace (2004); De Wet (2013); Skorupka, D. and Kowacka, M. (2016); Lopes, Braga and Resende Filho (2015); Glória Junior (2014); Khan (2010); El-Sayegh (2008); Choudhry, Aslam and Arain (2014); Lee and Schaufeberger(2014); Neves, Da Silva, Salomon et al. (2014); El-Sayegh and Mansour (2015); Hwang, Zhao, See et al. (2015); Liu, Meng and Fellows (2015); Arashpour, Wakefield, Lee et al. (2016);Doubravsk`y, Doskoˇcil and Dohnal (2016); Mhatre, Thakkar and Maiti (2017); Zavadskas et al. (2010); Zayed et al. (2008); Bunni (2003); Mhatre, Thakkar and Maiti (2017); Iyer and Jha (2005); Mhatre, Thakkar and Maiti (2017); Zeng et al. (2007); Jiang and Klein (2000); Gholami (2012)

Selladurai Pitchaimuthu, Jitesh J. Thakkar and P.R.C. Gopal (2019); Tah, J.H.M., Thorpe, A. and cCaffer, R. (1993); Wideman, M.R. (1992); Mhatre, Thakkar and Maiti (2017); Wang and Yuan (2011); Boehm-(1991); Jiang and Klein (2000); Glória Junior-(2014); Schmidt et al. (2001); Wallace-(2004); Khan (2010);Buckl et al. (2011); Lamersdorf (2011); De Wet (2013);El Emam (2008); Pinna and Arakaki (2009); Gholami (2012); Nakashima and Carvalho (2004); Iyer and Jha (2005); Lopes, Braga and Resende Filho (2015); Boehm (1991); Thamhain (2013); Neves, Da Silva, Salomon et al.(2014); Rao, Zhang, Shi et al. (2014);Liu, Meng and Fellows (2015); Yim, Castaneda, Doolen et al. (2015); Arashpour, Wakefield, Lee et al. (2016); Bozorg-Haddad, Orouji, Mohammad-Azari et al. (2016); Arena, Azzone, Cagno et al. (2014);Choudhry, Aslam and Arain (2014); Rao, Zhang, Shi et al. (2014); El-Sayegh and Mansour (2015); Elzamly and Hussin (2015);Arashpour, Wakefield, Lee et al. (2016); Liu and Yuliani (2016); Perrenoud, Smithwick, Hurtado et al. (2016); Rezazadeh Kmani,Moeni (2016);Bannerman (2008); Chang Lee, Lee and Li (2009); Dey, Clegg and Bennett (2010); Hu, Zhang, Ngai et al. (2013); Lopes, Braga and Resende Filho (2015)

Pitchaimuthu, Thakkar and Gopal (2019); Wideman (1992); Mhatre, Thakkar and Maiti (2017); Zavadskas et al. (2010); Zayed et al. (2008); Glória Junior (2014); Bunni (2003); Boehm (1991); El Emam (2008); Khan (2010); Gholami (2012); De Wet (2013); Skorupka, D. and Kowacka, M. (2016); Pinna and Arakaki (2009); Verner (2014); Tang, Qiang, Duffield et al. (2007); Hwang, Zhao, See et al. (2015); Yim, Castaneda, Doolen et al. (2015); Liu and Yuliani, (2016);

Authors

(continued)

11,88

8,43

8,43

% Relevance

Table 3. Risk factors identified with the respective authors, grouped by the areas of knowledge of the PMBOK® and percentage of relevance

110 E. C. Marques et al.

Risk

Market inflation; exchange rate fluctuation; deviations in costs and excessive expenses; public credit; inflation, public opposition; financing risk; overhead of operating costs; risk of payment of expenses; price change; waste / non-availability of labor; economy; estimating bad costs; and budget, bidding and financial problems; change in tax regulation

Quality problems; measurement data errors; changes in documentation; unpreparedness for measurement; execution incompatible with measurements; lack of system documentation; failure to identify technical needs; lack of infrastructure documentation; and documenting and managing processes

Lack of competence; team members; engineering experience; auditors; project manager’s commitment; professional knowledge; insufficient or inappropriate personnel involved; lack of knowledge/competence of those involved in the project; unmotivated team; inability to manage projects; personal turnover; failure in knowledge management; lack of commitment; insufficient staff; lack of maturity of the development team; lack of confidence; turn-over; constant adaptation of the team; cultural barriers of the project team; team production capacity / team size; and team skills / experience / efficiency; failure of human resources; change in legislation; Complex tasks

Communication failures; failure to identify the communication format; and inadequate communication

Knowledge areas

Costs Management

Quality Management

Project Resource Management

Communications Management

Selladurai Pitchaimuthu, Jitesh J. Thakkar and P.R.C. Gopal; Wideman, M.R. (1992); Mhatre, Thakkar and Maiti (2017); Wang and Yuan (2011) Glória Junior (2014); Wallace (2004); Khan (2010); El Emam (2008); Gholami (2012); Pinna and Arakaki (2009) Tang, Qiang, Duffield et al. (2007);Hu, Zhang, Ngai et al. (2013);Thamhain (2013) Choudhry, Aslam and Arain (2014); Arashpour, Wakefield, Lee et al. (2016); Liu and Yuliani (2016);

Selladurai Pitchaimuthu, Jitesh J. Thakkar and P.R.C. Gopal (2019); Skorupka, D. and Kowacka, M. (2016); Wideman, M.R. (1992); Gloria Junior (2014) Boehm (1991); Jiang and Klein (2000); Glória Junior (2014); Schmidt et al. (2001); Wallace (2004); Bannerman (2007); El Emam (2008); Khan (2010);Lamersdorf (2011); Gholami (2012); De Wet (2013); Mhatre, Thakkar and Maiti (2017); Zavadskas et al. (2010); Zayed et al. (2008); Bunni (2003); Wang and Yuan (2011); Iyer and Jha (2005); Zeng et al. (2007); Lopes, Braga and Resende Filho (2015); Pinna and Arakaki (2009); Buckl et al. (2011); Tang, Qiang, Duffield et al.(2007); Thamhain (2013); Neves, Da Silva, Salomon et al. (2014); El-Sayegh and Mansour(2015);Doubravsk`y, Doskoˇcil and Dohnal (2016); Hwang (2016); Xu, Y., Chan, A.P., Xia, B., Qian, Q.K., Liu, Y. and Peng, Y. (2015);

Selladurai Pitchaimuthu, Jitesh J. Thakkar and P.R.C. Gopal (2019);Wideman, M.R. (1992); Mhatre, Thakkar and Maiti (2017); Zavadskas et al. (2010); Zayed et al. (2008); Glória Junior (2014); Bunni (2003); Boehm (1991); El Emam (2008); Khan (2010); Gholami (2012); De Wet (2013); Skorupka, D. and Kowacka, M. (2016);Pinna and Arakaki (2009); Verner (2014); Tang,Qiang, Dfield et al. (2007);Hwang, Zhao, See et al. (2015); Yim, Castaneda, Doolen et al. (2015); Liu and Yuliani, (2016)

Selladurai Pitchaimuthu,Jitesh J. Thakkar and P.R.C. Gopal (2019); Tah, J.H.M., Thorpe, A. and McCaffer, R. (1993); Wideman, M.R. (1992); Xu, Y., Chan, A.P., Xia, B., Qian, Q.K., Liu, Y. and Peng, Y. (2015); Mhatre, Thakkar and Maiti (2017); Wang and Yuan (2011); Lopes, Braga and Resende Filho(2015); Choudhry, Aslam and Arain (2014); Elzamly and Hussin (2015); Rodney, Ducq,Breysse et al. (2015); Doubravsk`y, Doskoˇcil and Dohnal (2016); Guerrero-Liquet, Sánchez-Lozano, García-Cascales et al. (2016); Liu and Yulani (2016); Stojanovi´c (2016); Ahmadi, Behzadian, Ardeshir et al.(2017)

Authors

Table 3. (continued)

(continued)

6,51

12,26

7,66

6,13

% Relevance

A Systematic Review of Risk Management 111

Risk

Natural disasters; failure to effectively manage project risk; technical risk; government intervention; nationalization / expropriation; change in market demand; legislative developments; Changes in the organization; Competition; Physical security; Rules and Regulation; New or unknown subject; Changes in project ownership or senior management; Complex systems; Lack of mature / existing technologies; Execution failure in real time; Failure in risk management; Technical novelty under development; Technical failure of development; System test failure; Technical infrastructure failure; Lack of contingencies; Immature technology; Government bureaucracy / law / rules and regulations; Customers or demand risk; Environmental risk; Necessary technology; Risks related to quality; Company reputation; Contingency risk; Safety; Storage / stock capacity

Failure in contracts and acquisitions; project with several suppliers; failure to manage third parties; problems with third party technical artifacts; contracts; and supplier; defective equipment; technical failures

Knowledge areas

Risk Management

Project Procurement Management

Selladurai Pitchaimuthu, Jitesh J. Thakkar and P.R.C. Gopal (2019); Wideman, M.R. (1992); Lopes, Braga and Resende Filho (2015); Glória Junior (2014);Schmidt et al. (2001); Khan (2010); Lamersdorf (2011); De Wet (2013); Pinna and Arakaki (2009); Boehm (1991); El Emam (2008); Tang, Qiang, Duffield et al. (2007); Hu, Zhang, Ngai et al. (2013); Thamhain (2013);Liu, Meng and Fellows (2015); Yim, Castaneda, Doolen et al. (2015); Arashpour, Wakefield, Lee et al. (2016); Liu and Yuliani (2016); El-Sayegh (2008); Chang (2015);Hwang, Zhao, See et al. (2015); Roney, Ducq, Breysse et al. (2015); Rezazadeh Kermani and Momeni (2016); Jamshidi, Ait-Kadi and Ruiz (2017)

Selladurai Pitchaimuthu; Jitesh J. Thakkar and P.R.C. Gopal (2019); Tah, J.H.M., Thorpe, A. and McCaffer, R. (1993); Wideman, M.R. (1992);Skorupka, D. and Kowacka, M. (2016); Wideman, M.R. (1992); Wang and Yuan (2011); Xu, Chan, Xia, Qian, Liu and Peng, Q.K., Liu, Y. and Peng, Y. (2015); Glória Junior (2014); Schmidt et al. (2001);Bannerman (2007); Wallace (2004); Khan (2010); El Emam (2008); Renuka et al. (2014); Lopes, Braga and Resende Filho(2015); Lamersdorf (2011); Pinna and Arakaki (2009); Tang, Qiang, Duffield et al.(2007);Thamhain(2013); Choudhry, Aslam and Arain (2014); Lee and Schaufelberger (2014); Rao, Zhang, Shi et al.(2014); El-Sayegh and Mansour (2015);Chang Lee, Lee and Li (2009); Tsai and Yang (2009); Arashpour, Wakefield, Lee et al. (2016); Liu and Yuliani (2016); Perrenoud, Smithwick, Hurtado et al.(2016); Choudhry, Aslam and Arain(2014); El-Sayegh and Mansour (2015); Hwang, Zhao, See et al. (2015); Liu, Meng and Fellows (2015); Guerrero-Liquet, Sánchez-Lozano, García-Cascales et al. (2016); Rezazadeh Kermani and Momeni (2016); Hu, Zhang, Ngai et al. (2013); Arena, Azzone, Cagno et al. (2014); Fang and Marle (2012); Jamshidi, Ait-Kadi and Ruiz (2017); Lu and Yan (2013); Galli (2017); and Rodney, Ducq, Breysse et al. (2015)

Authors

Table 3. (continued)

(continued)

9,58

16,09

% Relevance

112 E. C. Marques et al.

Project Stakeholder Management

Source: from the authors

Total:

Risk

Political developments; coordination among stakeholders; lack of adequate user involvement; lack of powers for project management; conflict between user departments; failure to manage users’ final expectations; lack of cooperation from users; failure to obtain customer commitment; inadequate definition of roles and responsibilities; lack of commitment from top management; failure to manage expectations; inability to create commitment to the user; conflicts; culture; project leadership; and stakeholders

Knowledge areas Selladurai Pitchaimuthu; Jitesh J. Thakkar and P.R.C. Gopal(2019); Tah, J.H.M., Thorpe, A. and McCaffer, R. (1993); Mhatre, Thakkar and Maiti (2017); Zavadskas et al. (2010); Zayed et al. (2008); Bunni (2003); Zeng et al. (2007); Glória Junior (2014); Nakashima and Carvalho (2004); Schmidt et al. (2001); Buckl et al. (2011); Bannerman (2007); Jiang and Klein (2000); Wallace (2004); Lopes, Braga and Resende Filho (2015); Boehm (1991); El Emam (2008); Bannerman (2008); Chang Lee, Lee and Li (2009); Tsai and Yang (2009); Arashpour, Wakefield, Lee et al. (2016); Liu and Yuliani (2016); Perrenoud, Smithwick, Hurtado et al.(2016); El-Sayegh and Mansour (2015); Jamshidi, Ait-Kadi and Ruiz (2017); Arena, Azzone, Cagno et al. (2014); Choudhry, Aslam and Arain (2014); Neves, Da Silva, Salomon et al. (2014); Elzamly and Hussin (2015); Rezazadeh Kermani and Moomeni (2016); Firmenich (2017); Hwang, Zhao, See et al. (2015); and Doubravsk`y, Doskoˇcil and Dohnal (2016)

Authors

Table 3. (continued)

100

13,03

% Relevance

A Systematic Review of Risk Management 113

114

E. C. Marques et al.

within a limited timeframe is divided into activities associated with risks. These risks result in additional execution time and cost overhead. Some tools also allow managers to increase the resources allocated to an activity, in addition to reducing the duration of the project. So, they are not integrated with all the activities of the project. Table 3 shows the 149 risk factors identified in the 77 articles surveyed, with respective authors and were grouped into the 10 PMBOK® knowledge areas [49]. The risk factors were classified by the percentage of authors who cited them, considering the most relevant risk factors, those with the most citations, thus showing the importance of prioritizing immediate risk reduction / mitigation action by the project management team. The PMBOK® guide, in its sixth edition [49], presents the following 10 areas of knowledge: Project Integration Management; Scope Management; Schedule Management; Cost Management; Quality Management; Project Resource Management; Communications Management; Risk management; Project Procurement Management and Project Stakeholder Management.

5 Conclusion The research carried out allowed us to conclude that System Dynamics is a method that enables the Project Risk Management process, providing support to the decision process. The main contribution of this article is the identification of 149 risk factors, which were grouped into the 10 areas of knowledge of the PMBOK, which present a high severity, which need immediate risk reduction/mitigation action by the risk management team. Projects. They can be useful for future studies for anticipated project risks. The areas that need attention are: Risk Management, Stakeholder Management, Project Resource Management and Schedule Management, which account for 53.26% of the risk factors in projects and therefore appear as the risk factors. Most relevant risks. The application of System Dynamics was evidenced in IT projects related to software development and cloud computing, showing that it quickly changed the conventional way in which IT products and services are delivered. System Dynamics was also successfully applied to understand the effect of the involvement provided by the management team in the risk reduction process, the results of which helped the project management team to identify the highly serious risk factors that needed immediate reduction action. / risk mitigation, proving to be an efficient Risk Management method. Considering the dynamic nature and complexity of project management, the Systems Dynamics approach has been successfully employed. Future studies are recommended involving the creation of risk contingency plans to be executed as a flexibility to manage uncertainties and risks that can deviate the project from its original objectives of time, cost and quality and there are opportunities for empirical studies in several sectors, as most of the empirical studies are concentrated in the area of civil construction and information technology.

References 1. Lyneis, J.M., Coopera, K.G., Elsa, S.A.: Strategic management of complex projects: a case study using system dynamics - Syst. Dyn. Rev. 17, 237–260 (2001). https://onlinelibrary. wiley.com/resolve/doi?DOI=10.1002/sdr.213

A Systematic Review of Risk Management

115

2. Gomes, O.O.: Dinâmica de Sistemas como Método de Gestão de Riscos: Aplicação na Gestão de Riscos em Recursos Hídricos / Oseias Oliveira Gomes; orientador Ricardo Matos Chaim. – Brasília (2018). https://repositorio.unb.br/handle/10482/34150 3. de Sousa, M.R., Ribeiro, A.L.P.: Systematic review and meta-analysis of diagnostic and prognostic studies: a tutorial Arq. Bras. Cardiol. vol.92 no. 3 São Paulo march (2009). https:// doi.org/10.1590/S0066-782X2009000300013 4. Sampaio, R.F., Mancini, M.C.: Systematic review studies: a guide for careful synthesis of scientific evidence. Rev. Bras. Physiotherapy 11(1) (2007). https://doi.org/10.1590/S141335552007000100013 5. Lyneis, J.M., Ford, D.N.: System dynamics applied to project management: a survey, assessment, and directions for future research. Syst. Dyn. Rev. 23(2/3) (2007). Summer/Fall 2007 - Publicado online na Wiley InterScience (www.interscience.wiley.com) https://doi.org/10. 1002/sdr.377 6. Bernstein, P.L.: Desafio aos deuses: A fascinante história do Risco. Editora Alta Books (2018). ISBN 978–8550802732 7. PMI - Practice Standard for Project Risk Management (PMBOK®). Four Edition (English) Project Management Institute, 2009. ©2009 Project Management Institute, Inc. ISBN: 978– 1–933890–38–8. 2009 8. Forrester, J.W.: System Dynamics and the Lessons of 35 Years, Relatório Técnico D4224–4. MIT System Dynamics Group, Cambridge (1991) 9. Sterman, J.D.: Business dynamics: systems thinking and modeling for a complex world. McGraw-Hill Higher Education, New York (2000). ISBN 007238915X 10. Goh, Y.M., Love, P.E.D., Brown, H., Spickett, J.: “Organizational accidents: a systemic model of production versus protection. J. Manage. Stud. 49(1), 52–76 (2012). https://doi.org/10. 1111/j.1467-6486.2010.00959.x 11. Morecroft, J.D.W.: Strategic Modelling and Business Dynamics: A Feedback Systems Approach, 2nd edn. Wiley (2015). https://doi.org/10.1002/9781119176831 12. Mahato, B.K., Ogunlana, S.O.: Conflict dynamics in a dam construction project: a case study. Built Environment Project Asset Manage. 1(2), 176–194. https://doi.org/10.1108/204412411 11180424 13. Ogunlana, S.O., Li, H., Sukhera, F.A.: System dynamics approach to exploring performance enhancement in a construction organization. J. Constr. Eng. Manage. 129(5), 528–536 (2003). https://doi.org/10.1061/(ASCE)0733-9364(2003)129:5(528) 14. Hsu, P.-F., Ray, S., Li-Hsieh, Y.-Y.: Examining cloud computing adoption intention, pricing mechanism, and deployment model. Int. J. Inf. Manage. 34(4), 474–488 (2014). https://doi. org/10.1016/j.ijinfomgt.2014.04.006 15. Walterbusch, M., Martens, B., Teuteberg, F.: Evaluating cloud computing services from a total cost of ownership perspective. Manage. Res. Rev. 36(6), 613–638 (2013). https://www. emerald.com/insight/content/doi/10.1108/01409171311325769/full/html 16. PMI. Guia do Conhecimento em Gerenciamento de Projetos (GUIA PMBOK®). Fifth Edition (English Project Management Institute (2013). ISBN 978193589679 17. Zwikael, O., Ahn, M.: A Eficácia da Gestão de Riscos: Uma Análise do Projeto Risk Planejamento entre Indústrias e Países. Análise de Risco 31(1), 25–37 (2011). https://doi.org/10. 1111/j.1539-6924.2010.01470.x 18. Sanchez, H., et al.: A avaliaçãode risco aplicada a projetos, programas e portfólios. Int. J. Manag. Projects Bus. 2(1), 14–35 (2009). https://doi.org/10.1108/17538370910930491 19. Fang, C.; Marle, F.: Um modelo de rede de risco baseado em simulação para suporte de decisão em gerenciamento de risco de projetos. Sistemas de Apoio à Decisão 52(3), 635–644 (2012). https://doi.org/10.1016/j.dss.2011.10.021

116

E. C. Marques et al.

20. Veiga, E.S.D, Silva, E.M.: A systematic review of risk management in projects. Revista Produção Online. Florianópolis, SC, 20(3), 837–857 (2020). Associacao Brasileira de Engenharia de Producao - ABEPRO. https://doi.org/10.14488/1676-1901.v20i3.3636 21. Thamhain, H.: Gerenciando Riscos em Projetos Complexos. Project Manage. J. 44(2), 20–35 (2013). https://doi.org/10.1002/pmj.21325 22. Kerzner, H., Saladis, F.P.: Gerenciamento avançado do portfólio de projetos e o PMO: Multiplicando o ROI em velocidade de dobra. John Wiley & Sons Inc, Hoboken (2009) 23. Wang, M.T., Chou, H.Y.: Alocação de risco e tratamento de riscos de projetos rodoviários em Taiwan. J. Manage. Eng. 19(2), 60–68 (2003). https://doi.org/10.1061/(ASCE)0742-597 X(2003)19:2(60) 24. Shenhar, A.J.: Um tamanho não se encaixa em todos os projetos: Explorando domínios clássicos de contingência. Ciência da Gestão 47(3), 394–414 (2001). https://doi.org/10.1287/mnsc. 47.3.394.9772 25. Marcelino-Sádaba, S., et al.: Metodologia de gerenciamento de riscos de projetos para pequenas empresas. Int. J. Project Manage. 32(2), 327–340 (2014). https://doi.org/10.1016/j.ijp roman.2013.05.009 26. Levitt, R.E., Thomsen, J., Christiansen, T.R., Kunz, J.C., Jin, Y., Nass, C.: Simulando processos de trabalho de projetos e organizações: Em direção a uma teoria de micro-contingência do design organizacional. Ciência da Gestão 45(11), 1479–1495 (1999). https://doi.org/10. 1287/mnsc.45.11.1479 27. Ackermann, F., Eden, C., Williams, T.: Modelagem para litígio: Mistura de abordagens qualitativas e quantitativas. Interfaces 27(2), 48–65 (1997). https://doi.org/10.1287/inte.27. 2.48 28. Eden, C.E., Williams, T.M., Ackermann, F.A.: Desmantelamento da curva de aprendizagem: O papel das interrupções no planejamento de projetos de desenvolvimento. Int. J. Project Manage. 16(3), 131–138 (1998). https://doi.org/10.1016/S0263-7863(97)00053-7 29. Eden, C.E., Williams, T.M., Ackermann, F.A., Howick, S.: Sobre a natureza de disrupção e atraso (D&D) em grandes projetos. Revista da Sociedade de Pesquisa Operacional 51(3), 291–300 (2000). https://doi.org/10.1057/palgrave.jors.2600919 30. Gil, A.C.: Métodos e Técnicas de Pesquisa Social. 7. ed. Atlas, São Paulo (2019). ISBN 978–85–970–2098–4 31. Lakatos, E.M., Marconi, M.A.: Fundamentos de Metodologia Científica. 8. ed. Atlas, São Paulo (2017). ISBN 978–85–970–1076–3 32. Lyneis, J.M., Ford, D.N.: System dynamics applied to project management: a survey, assessment, and directions for future research. Syst. Dyn. Rev. [S.L.], 23(2–3), 157–189 (2007). Wiley. https://doi.org/10.1002/sdr.377. Disponível em: /https://onlinelibrary.wiley.com/doi/ abs/10.1002/sdr.377. Acesso em: 08 Nov 2020 33. Lyneis, J.M., Cooper, K.G., Els, S.A.: Strategic management of complex projects: a case study using system dynamics. Syst. Dyn. Rev. [S.L.], 17(3), 237–260 (2001). Wiley. https://doi.org/10.1002/sdr.213. Disponível em: https://onlinelibrary.wiley.com/doi/abs/10. 1002/sdr.213. Acesso em: 08 Nov. 2020. 34. Martins Júnior, J.: Como Escrever Trabalhos de Conclusão de Curso. 7. ed. Vozes, Petrópolis (2013). ISBN 978–85–326–3603–4 35. Sterman, J.D.: System Dynamics Modeling for Project Management. 1992. Disponível em: https://web.mit.edu/jsterman/www/SDG/project.pdf. Acesso em: 09 Nov 2020 36. Rodney, E., et al.: Uma abordagem de gestão integrada dos riscos do projeto e do projeto. IFACPapersOnLine 48(3), 535–540 (2015). https://doi.org/10.1016/j.ifacol.2015.06.136 37. Francês, S., Maule, J., Papamichail, N.: Comportamento de decisão, análise e suporte. Cambridge University Press, Cambridge (2009). https://doi.org/10.1017/CBO978051160 9947

A Systematic Review of Risk Management

117

38. Breysse, D., et al.: Gestão de riscos de projetos em projetos de construção: desenvolvimento de ferramentas de modelagem para favorecer uma abordagem multidisciplinar. Em: Segurança, Confiabilidade, Risco e Ciclo de Vida Performance de estruturas e infraestruturas Safety, Reliability, Risk and Life-Cycle Per. Deodatis, Ellingwood e Frangopol (2013). ISBN 978– 1–315–88488–2 39. Grimaldi, S., e al.: Uma estrutura para selecionar técnicas que suportem o gerenciamento de riscos de projetos. InTech, Rijeka (2012). https://doi.org/10.5772/50991 40. Bannerman, P.L.: Gestão de riscos e riscos em projetos de software: Uma reavaliação. J. Syst. Softw. 81(12), 2118–2133 (2008). https://doi.org/10.1016/j.j.jss.2008.03.059 41. Meredith, J.R.: Theory Building através de Métodos Conceituais. Int. J. Oper. Prod. Manage. 13(5), 3–11 (1993). https://doi.org/10.1108/01443579310028120 42. Raiffa, H.: A orientação prescritiva da tomada de decisão: uma síntese deysis decisório, tomada de decisão comportamental e teoria dos jogos. In: Ríos, S. (ed.). Teoria da Decisão e Análise de Decisões: Tendências e Desafios. Norwell Kluwer Academic Publishers (1994). https://doi.org/10.1007/978-94-011-1372-4_1 43. Bertrand, J.W.M., Fransoo, J.C.: Operations management research methodgies using quantitative modeling. Revista Internacional de Operações e Gestão de Produção 22(2), 241–264 (2002). https://doi.org/10.1108/01443570210414338 44. Da Nóbrega, D.A., Fenner, G., Lima, A.S.: Uma proposta de metodologia de Gestão de Riscos para projetos de tecnologia da informação. IEEE Lat. Am. Trans. 12(4), 643–656 (2014). https://doi.org/10.1109/TLA.2014.6868866 45. Forrester, J.W.: Industrial dynamics. J. Oper. Res. Soc. 48(10), 1037–1041 (1997). https:// doi.org/10.1057/palgrave.jors.2600946 46. Ghosh, A.: Dynamic Systems for Everyone. Springer, Cham (2017). https://doi.org/10.1007/ 978-3-319-43943-3 47. Sterman, J.D., Bayer, S.: Systems thinking and modeling for a complex world. Management (2000). https://doi.org/10.1108/13673270210417646 48. Meier, H., Boßlau, M.: Design and engineering of dynamic business models for industrial product-service systems. In: The Philosopher’s Stone for Sustainability, pp. 179–184. Springer, Heidelberg (2013). https://doi.org/10.1007/978-3-642-32847-3_30 49. Um Guia do Conhecimento em Gerenciamento de Projetos (Guia PMBOK)/Project Management Institute - Sexta edição. | Newtown Square, PA: Project Management Institute, 2017. | Série: Guia PMBOK |©2017 Todos os direitos reservados ISBN: 9781628253924 50. Pitchaimuthu, S., Thakkar, J.J., Gopal, P.R.C.: Modelling of risk factors for defence aircraft industry using interpretive structural modelling, interpretive ranking process and system dynamics. Measur. Bus. Excell. [S.L.] 23(3), 217–239 (2019). Emerald. https://doi. org/10.1108/mbe-05-2018-0028. Disponível em: https://doi-org.ez54.periodicos.capes.gov. br/10.1108/MBE-05-2018-0028. Acesso em: 21 Nov 2020 51. Xu, Y., Chan, A.P., Xia, B., Qian, Q.K., Liu, Y., Peng, Y.: Fatores críticos de risco que afetam a implementação de projetos de desperdício de energia PPP na China. Energia Aplicada 158, 403-411 (2015) 52. Wideman, M.R.: Gerenciamento de Riscos de Projetos e Programas: Um Guia para Gerenciar Riscos e Oportunidadesde Projetos. Project ManagementInstitute, Newtown Square, PA (1992) 53. Tah, J.H.M., Thorpe, A., McCaffer, R.: Projeto empreiteiro arrisca a alocação de contingência nosaproximação linguística. Sistemas de Computação em Engenharia 4(2/3), 281–293 (1993) 54. Skorupka, D., Kowacka, M.: Identificação do fator de risco de desenvolvimento de umaoperação de estrada no trabalho de levantamento de luz. Arquivo de Engenharia Civil 60(2), 183–190 (2016)

118

E. C. Marques et al.

55. Yan, M.-R.; Lee, Y.-H.: Measuring project performance and dynamic risk management for engineering design alliance. Measur. Bus. Excell. [S.L.], 1–22 (2020). Emerald. https://doi. org/10.1108/mbe-04-2020-0054. Disponível em: https://www.emerald.com/insight/content/ doi/10.1108/MBE-04-2020-0054/full/html. Acesso em: 21 Nov 2020 56. Mhatre, T.N., Thakkar, J.J., Maiti, J.: Modelling critical risk factors for Indian construction project using interpretive ranking process (IRP) and system dynamics (SD). Int. J. Qual. Reliabil. Manage. [S.L.], 34(9), 1451–1473 (2017). Emerald. https://doi.org/10.1108/ijqrm09-2015-0140. Disponível em: www.emeraldinsight.com/0265-671X.htm. Acesso em: 21 Nov 2020 57. Lopes, J.S., Braga, J.L., Resende Filho, M.A.: Systems dynamics model for decision support in risk assessment in software projects. J. Software: Evol. Process, [S.L.] 27(12), 976– 989 (2015). Wiley. https://doi.org/10.1002/smr.1754.Disponívelem: https://doi-org.ez54.per iodicos.capes.gov.br/10.1002/smr.1754. Acesso em: 21 Nov 2020 58. Boateng, P., Ahiaga-Dagbui, D., Chen, Z., Ogunlana, S.: Modelling Economic Risks in Megaproject Construction: A systemic Approach. In: Proceedings of the 31St Annual Association of Researchers in Construction Management Conference, Arcom 2015, Lincoln, 1(1), 115–124 (2015). Disponível em: https://www.arcom.ac.uk/-docs/archive/2015ARCOM-Full-Proceedings-Vol-1.pdf. Acesso em: 21 Nov 2020 59. Cubin, L., Yunqi, L., Shuke, L.: Human resources risk element transmission model of construction project based on system dynamic. Open Cybern. Syst. J. [S.L.], 9(1), 295–305 (2015). Bentham Science Publishers Ltd.. https://doi.org/10.2174/1874110x01509010295. Disponível em: https://benthamopen.com/ABSTRACT/TOCSJ-9-295. Acesso em: 21 Nov 2020.

A Brokering-Based Interaction Protocol for Dynamic Resource Allocation in Steel Production Processes Vincenzo Iannino(B)

, Claudio Mocci, and Valentina Colla

Scuola Superiore Sant’Anna, TeCIP Institute, Via Moruzzi 1, 56124 Pisa, Italy {v.iannino,c.mocci,v.colla}@santannapisa.it

Abstract. Steel manufacturing involves a series of dynamic processes that require an efficient and effective management of the plant resources. The coordination and the allocation of the resources is an important aspect for ensuring a continuous production flow as well as a good quality of the final products, especially when unexpected events can compromise the overall system performance. This paper presents an agent-based protocol for dynamic resource allocation in order to establish collaboration among agents within steel production processes. The proposed protocol is based on the brokering mechanism and is designed in order to solve the problem of the concurrency, which arises when several agents are interested in using the same resources, and to handle dynamic changes, such as unexpected events that can affect the resource allocation process. Experimental results show how the designed protocol allows the agents coordination by guaranteeing the use of the resources and the correct flow of the production. Keywords: Multi-agent systems · Cooperation · Dynamic resource allocation · Negotiation · Brokering interaction protocol · Steel production · Factory of the future

1 Introduction The steel industry is active in all parts of the world and its production processes have a significant impact on global economy [1]. Melting, refining, casting, forming (e.g. hot and cold rolling) and coating are examples of steel production processes, and each production step needs multiple resources, such as materials, machineries, transport systems (e.g. cranes, forklifts), etc. Allocation and efficient exploitation of plant resources is an important aspect that allows maximization of productivity by reducing waste of materials and production costs. The strong acceleration of the digitalization promoted by Industry 4.0 is transforming the steel sector [2] and the new generation of steel manufacturing processes is pushed towards a new vision of the production systems, the so-called Factory of the Future (FOF) [3]. In the FOF, the plant structure is more flexible and digitalized, processes are dynamic and capable to self-adapt to continuous changings that can affect the production flow. In this context, classical optimization techniques based on linear programming © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 119–129, 2021. https://doi.org/10.1007/978-3-030-72654-6_12

120

V. Iannino et al.

algorithms, which are used to solve different types of resource allocation problems are less effective compared to agent-based approaches [4]. In agent-based approaches, a community of distributed intelligent agents interact in order to solve problems that are beyond their individual capabilities. Agent-based technologies provide efficient and interesting solutions to a wide range of problems in different areas of interest such as, for instance, manufacturing and process industry [5–7], post-earthquake operations [8], traffic control [9], smart grids [10]. Thus, Multi-Agent Systems (MAS) represent a suitable approach to embrace the FOF vision thanks to their inherits features like decentralization, intelligence, autonomy, and adaptability. An important aspect of MAS is the coordination. The are several reasons why multiple agents need to be coordinated: for instance, agents’ goals may cause conflicts among agents’ actions or agents may have different capabilities and knowledge. Resource allocation, which is the process of distributing a number of resources among a set of agents [11], is an activity that require agents to cooperate and coordinate their actions. It can be realized through centralized, semi-centralized or decentralized approaches [12]: centralized approaches are based on a hierarchical organization of the system; semi-centralized approaches exploit special functional agents, the so-called Middle Agents (MAs), holding the function of brokers [13] or recruiters [14]; decentralized approaches are mainly based on bidding. This last technique includes the well-known Contract Net Protocol (CNP) proposed by Smith [15] and its variants versions [16,17], auction-based and alternating offer-based protocols [18]. Each of the aforementioned techniques show advantages and limitations, establishing what is the best technique is hard, and this research area is a continuous challenge for researchers. In this paper, a protocol for resource allocation in steel production processes is proposed, which is based on brokers in MAS. Brokers are specialized agents that coordinate the resource allocation by dynamically centralizing the demands of client agents and the resources offered by service agents. The use of brokering mechanism has been selected, as it is widely adopted in agent-based concurrent design and manufacturing systems and ensures the coordination of different groups of agents. The proposed protocol is designed to deal with concurrent negotiations with multiple resources and to handle system dynamic changes caused by anomalies that can affect the resource allocation process. In particular, the broker agents’ role consists in both updating the client agents concerning resources availability and ensuring an ordered access to resources. Furthermore, the protocol is designed in order to properly work even if no resources are available, e.g. due to maintenance operations. The paper is organized as follows: Sect. 2 describes the considered industrial problem. Section 3 presents the proposed brokering interaction protocol, briefly describes the agent-based model considered for the resource allocation problem and overviews the brokering mechanism. Section 4 presents the experiments of the designed protocol, while Sect. 5 provides some concluding remarks and hints for future work .

A Brokering-Based Interaction Protocol for Dynamic Resource

121

2 Steel Production Process Case Study The industrial problem considered in this work concerns the resource allocation of a part of a generic steel production process, namely the cold rolling process [19]. The production starts by loading hot-rolled strip stocks into the Pickling Line (PL) by using semi-automated cranes. The pickling process allows removing scales, which consist of thin layers of iron oxide crystals, from strips by pulling them through a series of tanks filled with corrosive solutions, typically hydrochloric acids. After pickling, the forming process takes place. The descaled strip is rolled through the Cold Rolling Line (CRL), which is usually a tandem mill of four to six stands. This process is very important, as it determines the hardness and other material properties of the finished product. After the cold rolling, in order to reduce the hardness of the strip and increase its ductility, a heat treatment by annealing is frequently applied. The annealing process consists in heating the rolled strip above its recrystallization temperature for a proper period at a suitable temperature before cooling. The Annealing Line (AL) can be continuous or in batch, namely it can process one strip at time or many strips contemporary. After annealing, the physical properties of the strip steel are improved through a Temper Mill (TM) and/or other treatments based on the quality required by the customers. The temper treatment is a rolling operation that gives the final surface to the strip. If required, a shearing line at the end of all process cuts strips into sheets. Depending on the type of plant, some process steps are continuously carried out without coiling the strips and using intermediate storages, but usually semi-finished products are stocked between subsequent stages. The considered cold rolling process is schematically depicted in Fig. 1. There are four stages, each of which composed of two parallel machines: the first stage has 2 pickling lines, the second stage has 2 cold rolling lines, the third stage has 2 continuous annealing lines, and the last stage has 2 temper mills. Two parallel machines may be different, but in this use case they are assumed to be identical. Strips are transported from one stage to another by means of a crane. In particular, there is a crane for each subsequent stage that takes a minimum time to transport the strips. Furthermore, semi-finished products are stocked between subsequent stages. Several products are usually processed during the day thus a proper allocation of the production resources is of utmost importance. Such allocation is usually done in order to optimize the company’s objectives and goals (e.g., maximizing plant utilization, minimizing costs, etc.) without violating any of the constraints. Nevertheless, unforeseen events may happen, such as breakdowns, special maintenance operations or delays in the arrival of raw materials, that can affect the normal production flow and, as a consequence, prevent fulfilment of the desired goals.

122

V. Iannino et al.

Fig. 1. Cold rolling process scheme

3 Brokering-Based Interaction Protocol for Dynamic Resource Allocation 3.1 Agent-Based Model The agent-based model considered in this paper to evaluate the proposed resource allocation protocol is composed of the following agents: • User Agents (UAs) represent the semi-finished products that must be worked and are interested in using plant resources. • Resource Agents (RAs) are responsible of the resources. There are two types of RAs: Machine Resource Agents (MRAs) and Crane Resource Agents (CRAs). In this work, the focus is mainly on the allocation of machines that represent fixed non-sharable resources, e.g., resources that are not mobile and can be allocated to one UA at time, and the allocation of cranes, which are mobile non-sharable resources. The status of a machine resource can be available, busy or in maintenance while the crane resource can be available or in maintenance. Each UA may prefer a resource over another based on some desired requirements and preferences. In this model, UAs want to use a resource as soon as possible, so the preference is on the first available resource, but other types of preferences can be specified, by extending the possible use cases. 3.2 Resource Allocation by Brokers The brokering mechanism basically consists of receiving a request message from an intelligent agent, understanding the request, finding suitable receptors for the message, and multicasting the message to the selected group of agents [12]. Thus, the resource allocation realized through the brokering mechanism involves three nodes namely, the UA, the Broker Agent (BA) and RAs, and is mainly composed of three phases: • Request: the UA searches for a resource to perform a specific operation and asks the BA to solve its problem.

A Brokering-Based Interaction Protocol for Dynamic Resource

123

• Search: the BA searches its database for suitable RAs that can perform the UA’s operation and interact with them. • Reply: the BA allocates the resource to the UA and forwards the result back to the UA. The use of broker agents can significantly simplify the interaction with agents in MAS. Brokers provide monitoring and notification functionalities, allow the system to be adaptable and robust in dynamic situations, supporting scalability and security control at the brokering agent. 3.3 Proposed Brokering-Based Interaction Protocol for Dynamic Resource Allocation The proposed protocol is based on the brokering mechanism and involves specialized middle agents with determined roles: • Keep track of the status of the resources in a stage of the system. Each RA automatically informs MAs when its status changes through specified messages (I[Available] and I[Maintenance]). • Allocate resources amongst UAs. The MA policy is based on receiving order of UAs’ requests (FIFO), but other policies can be used. • Advise UAs whether resources are unavailable through a specific message (I[Wait]). 3.3.1 Machine Resource Allocation For each production stage, a MA is set up and manages the requests of UAs passing through that production step. UAs and MRAs initially do not directly communicate with each other but exploit MAs. An UA knows only its working routing (production steps), whereas a MA knows the machines available within its own production step. The communication protocol is shown in Fig. 2. The associated communications are: • Resource Allocation Request (RAR): The UA sends a RAR to MA asking to be processed by a machine based on some specifications included into the message. In the considered model, the specification is referred to any machine available at the current production step and a description of the product parameters. • Processing Request (PR): The MA determines the potential machine resources based on the specifications included in the RAR and the status of the machine resources and forwards a PR to the suitable MRA attaching product details. • Resource Agreement (RA): The MRA replies with an agreement message that is binding for the resource itself, thus it changes its status to busy. • Resource Reply (RR): the MA updates the list of the available machine resources and sends a RR to the UA with the information of the machine resource that is in charge of working it.

124

V. Iannino et al.

Fig. 2. Brokering interaction protocol for machine resources allocation

3.3.2 Crane Resource Allocation While the machine resources allocation is carried out through a dedicated MA, as several UAs run concurrently and multiple MRAs might process the same product and manage several negotiations, the crane resources allocation is directly performed by MRAs themselves, being only one crane available at each production step. Thus, MRAs can take the role of MAs. In presence of multiple cranes at the same production stage dedicated MAs should be set as for the machine resources allocation. Therefore, the proposed schema for the crane resources allocation can also be seen as a particular case of brokering system: a MRA becomes a broker and manages the allocation of a single crane resource to a specified UA. Once a UA receives a RR from MA, it starts a broker interaction with the specific MRA as shown in Fig. 3 and the associated communications are the following: • Resource Allocation Request (RAR): The UA sends a RAR to MRA asking to be loaded by a crane into the machine based on some specification included into the message. • Processing Request (PR): The MRA checks the status of the crane resource and forwards a PR to the CRA at the current production step attaching product details. • Resource Agreement (RA): The CRA replies with an agreement message, stores all the MRAs’ requests in a queue and processes them in a FIFO order.

Fig. 3. Brokering interaction protocol for crane resources allocation

A Brokering-Based Interaction Protocol for Dynamic Resource

125

• Resource Reply (RR): the MRA sends a RR to the UA with the information of the crane resource that is in charge of transporting it.

3.3.3 Machine and Crane Resources Allocation Under Maintenance Nowadays, maintenance is extremely important for safety and reliability of manufacturing processes. Maintenance is typically of two types: run-to-failure or preventive [20]. Under maintenance operations, the entire production flow must not be compromised, and must be capable to dynamic reallocate the available resources. In order to handle maintenance events, the design protocol is extended with dedicated messages (I[Maintenance]) that are sent by RAs. If a machine undergoes maintenance operations, the MRA sends a message to all nodes that have a direct connection with it, i.e. the MA, the CRA and eventually the involved UA. The MA updates the status of the machine resource; the CRA stops processing the requests of the MRA, as the time required by the maintenance is not known a-priori, and the UA renegotiates the resource allocation with other MRAs through the MA mediation. An example of machine resources allocation under maintenance is reported in Fig. 4. Similar considerations can be done for the crane resources. As already explained in Sect. 2, each production step has one crane which moves the stocked semi-finished products. If a crane undergoes maintenance operations, no other crane can load products to machines and the production process is blocked. This situation is handled through messages sent by the CRA to all the connected MRAs, such as shown in Fig. 5. MRAs update the status of the crane resource waiting until CRA sends a message (I[Available]), which warns them that the crane is again available.

Fig. 4. Machine resources allocation maintenance management protocol

126

V. Iannino et al.

Fig. 5. Crane resources allocation maintenance management protocol

4 Results and Discussion In order to test the proposed protocol for dynamic resource allocation under dynamic changing conditions two scenarios have been considered: the former without unexpected events and the latter where plant resources are subjected to unexpected events such as breakdowns or maintenance operations that compromise the production flow. Both simulated scenarios consider the production of twenty hot-rolled strips P1–P20 with the same characteristics and quality. The machines average processing times used for the simulations are reported in Table 1. Machines setup times are neglected. The time needed by a crane to load a stocked strip is considered of about 2 min. The data for the experiments are extrapolated from the report in [21]. Table 1. Machine processing times PL1 , PL2

CRL1 , CRL2

AL1 , AL2

TM1 , TM2

5 min

7 min

20 min

10 min

Gantt charts have been used to show the results of the experiments since they provide a graphical representation of resources allocation over time and are frequently used for scheduling purposes. The Gantt chart of the first scenario is shown in Fig. 6. Here, no perturbations affect the production system, all the requests of the UAs are satisfied by MAs and as a result the global makespan is 4 h 25 min 53 s. The Gantt chart of the second scenario is depicted in Fig. 7. In the experiment reported here, different unexpected events are randomly generated on-line during the simulation. In deeper detail, a breakdown on one of the CRLs at stage two (CRL2 ) is simulated, the crane at stage 1 (Cr1 ) and the crane at stage 3 (Cr3 ) are temporary suspended by a special maintenance operation as well as one of the TMs at stage four (TM2 ). Despite the unexpected events of the production flow, a feasible resource allocation is provided by the proposed agent-based brokering mechanism showing a good flexibility and utilization of the plant resources reporting a makespan very close to the first scenario.

A Brokering-Based Interaction Protocol for Dynamic Resource

127

In both scenarios, the problem of the concurrency that rises when UAs are interested in using the same plant resources is solved through MAs mediation that coordinate the negotiation. UAs are informed by MAs once resources are available avoiding continuous interaction with them. This minimizes the cost of informing all UAs and the overall number of exchanged messages. Even if the policy adopted by MAs for the resources allocation is simple, it allows a feasible plant resources allocation also in perturbed situations that affect the production flow.

Fig. 6. Gantt chart of resources allocation without unexpected events

Fig. 7. Gantt chart of resources allocation with unexpected events

128

V. Iannino et al.

5 Conclusion The coordination and the allocation of the resources in steel manufacturing processes is an important aspect for ensuring a continuous production flow and good quality of the final products, especially when unexpected events can compromise the overall system performance. MASs represent suitable candidates for the new generation of steel smart factories where systems are fully connected, flexible and intelligent. In this work an agent-based brokering mechanism for plant resources allocation is proposed in order to establish collaboration among agents within steel production processes. The advantages of the propose protocol are robustness and adaptability in dynamic situations. The protocol avoids continuous polling of the resources by reducing both communication costs compared to decentralized resource allocation approaches and the probability of failure having a reference, which is the middle agent, which knows the status of the resources. Furthermore, the brokering protocol has been extended with a dedicated maintenance management protocol to deal with failures that can affect resources, by thus increasing its robustness. As a counterpart, the role of the MAs becomes central, with a load on it, which increases with the dimension and complexity of the considered system. Moreover, the whole system reliability depends on the reliability of the MAs. Future work will deal with the improvement of the protocol by exploiting the possibility of a communication among MAs and the use of predictive maintenance strategies in order to determine the actual mean-time-to-failure for each machinery for an optimal management of the stock materials. Furthermore, different policies to be adopted by MAs will be evaluated and tested, such as the ones based on metaheuristic approaches.

References 1. Word Steel Association: The Role of Steel Manufacturing in the Global Economy (2019) 2. Branca, T.A., Fornai, B., Colla, V., Murri, M.M., Streppa, E., Schröder, A.J.: The challenge of digitalization in the steel sector. Metals 10(2), 1–23 (2020) 3. Goncalves, R., Romero, D., Grilo, A.: Factories of the future: challenges and leading innovations in intelligent manufacturing. Int. J. Comp. Integrated Manuf. 30(1), 4–13 (2017) 4. Davidsson, P., Johansson, S.J., Persson, J.A., Wernstedt, F.: Agent-based approaches and classical optimization techniques for dynamic distributed resource allocation: a preliminary study. In: AAMAS 2003 Workshop on Representations and Approaches for Time-Critical Decent. Res./Role/Task Alloc., Melbourne, Australia (2003) 5. Monostori, L., Váncza, J., Kumara, S.: Agent-based systems for manufacturing. Ann. CIRP 55, 697-720 (2006) 6. Iannino, V., Mocci, C., Vannocci, M., Colla, V., Caputo, A., Ferraris, F.: An event-driven agent-based simulation model for industrial processes. Appl. Sci. 10(12), 1–22 (2020) 7. Iannino, V., Vannocci, M., Vannucci, M., Colla, V., Neuer, M.: A multi-agent approach for the self-optimization of steel production. Int. J. Simul. Syst. Sci. Technol. 19, 1–20 (2018) 8. Hooshangi, N., Alesheikh, A.A.: developing an agent-based simulation system for postearthquake operations in uncertainty conditions: a proposed method for collaboration among agents. ISPRS Int. J. Geo-Inf. 7(1), 1–22 (2018)

A Brokering-Based Interaction Protocol for Dynamic Resource

129

9. Vilarinho, C., Tavares, J.P., Rossetti, R.J.F.: A Conceptual MAS Model for Real-Time Traffic Control. In: Pereira, F., Machado, P., Costa, E., Cardoso, A. (eds.) Progress in Artificial Intelligence. EPIA 2015. Lecture Notes in Computer Science, vol. 9273. Springer, Cham (2015) 10. Singh, V.P., Kishor, N., Samuel, P.: Distributed multi-agent system based load frequency control for multi-area power system in smart grid. IEEE Trans. Ind. Electron. 64(6), 5151– 5160 (2017) 11. Chevaleyre, P.E., Dunne, U., Endriss, J., Lang, M., Lemaıtre, N., Maudet, J.A., Padget, S., Phelps, J., Rodrıguezz-Aguilar, A., Sousa, P.: Issues in multiagent resource allocation. Informatica 30(1), 3–31 (2006) 12. Shen, W.N., Douglas, H., Barthes, J.P.: Multi-Agent Systems for concurrent Intelligent Design and Manufacturing. CRC Press, Boca Raton (2014) 13. FIPA: FIPA brokering interaction protocol specification (2002). https://www.fipa.org/specs/ fipa00033/SC00033H.html. 14. Hudaib, A., Qasem, M.H., Obeid, N.: FIPA-based semi-centralized protocol for negotiation. In: Silhavy, R., Silhavy, P., Prokopova, Z. (eds) Cybernetics Approaches in Intelligent Systems. CoMeSySo 2017. Adv. Int. Syst. Comput., vol 661. Springer, Cham (2018) 15. Smith, R.: The contract net protocol: High level communication and control in distributed problem solver. IEEE Trans. Comput. 29, 1104–1113 (1980) 16. Aknine, S., Pinson, S., Shakun, M.F.: An extended multi-agent negotiation protocol. Auton. Agents Multi Agent Syst. 8(1), 5–45 (2004) 17. Dang, J., Huhns, M.N.: Concurrent multiple-issue negotiation for internet-based services. IEEE Internet Comput. 10, 42–49 (2006) 18. Adnan, M.H.M., et al.: Protocols for agent-based autonomous negotiations: a review. In: 3rd International Conference on Computer and Information Sciences (ICCOINS). IEEE (2016) 19. Roberts, W.L.: Cold Rolling of Steel. CRC Press, Boca Raton (1978) 20. Mobley, R.K.: An Introduction to Predictive Maintenance, 2nd edn. Butterworth-Heinemann, Burlington (2002) 21. Rentz, O., Jochum, R., Schultmann, F.: Report on Best Available Techniques (BAT) in the German Ferrous Metals Processing Industry. French German Institute for Environmental Research, Karlsruhe (1999)

A Tool to Support the Investigation and Visualization of Cyber and/or Physical Incidents Inês Macedo , Sinan Wanous , Nuno Oliveira(B) and Isabel Praça

, Orlando Sousa ,

School of Engineering from the Polytechnic of Porto (ISEP), Research Group on Intelligent Engineering and Computing for Advanced Innovation and Development (GECAD), Porto, Portugal {incar,sinai,nunal,oms,icp}@isep.pt

Abstract. Investigating efficiently the data collected from a system’s activity can help to detect malicious attempts and better understand the context behind past incident occurrences. Nowadays, several solutions can be used to monitor system activities to detect probable abnormalities and malfunctions. However, most of these systems overwhelm their users with vast amounts of information, making it harder for them to perceive incident occurrences and their context. Our approach combines a dynamic and intuitive user interface with Machine Learning forecasts to provide an intelligent investigation tool that facilitates the security operator’s work. Our system can also act as an enhanced and fully automated decision support mechanism that provides suggestions about possible incident occurrences. Keywords: Cyber and physical security · Data visualization · Investigation tool · Machine learning

1 Introduction Security investigation aims to explore the cause of an attack and how much it threatened the security of the targeted property. In case the security of a system is compromised, investigating over the data collected during the monitoring phase can bring important insights, both to improve detection and prevention, and to support mitigation and remediation strategies. Intrusion detection sensors can collect additional evidence, in the form of logs, related to system and network accesses, email services, file analysis, etc. In complex infrastructures, the number of systems can be huge, and thus the amount of data collected. The proper data visualization can be crucial to support the Security Operators in analyzing all the evidence collected. If we consider domains where both cyber and physical security monitoring solutions are used, not only the data volume is bigger, also the analyses of both domains are crucial to understand how complex attacks can be triggered, benefiting from the combination of cyber and physical threats. In this work, we describe an investigation tool, that is based on the time series visualization of cyber and physical attacks, and the way a machine learning and a visualization module are interconnected to support the detection and correlation of incidents. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 130–140, 2021. https://doi.org/10.1007/978-3-030-72654-6_13

A Tool to Support the Investigation

131

The paper is divided into multiple sections that can be detailed as follows. Section 2 provides an overview of related security frameworks, incident investigation research, and commercial security systems. Section 3 briefly describes the security ecosystem and our solution’s internal architecture. In Sect. 4, our system’s visualization component is illustrated with more detail, specifying the main functionalities as well as the most relevant dashboards that can be accessed. Finally, Sect. 5 provides a conclusion about how our approach facilitates the SOC Operator’s work and enhances the overall security process.

2 Related Work Authors of Thakur, K., et al. [1] describe crucial issues in cybersecurity domains. They focus on the issues of cyber security threats and summarizes the existing security models and research directions in each field. Moreover, a basic network infrastructure for applications and systems vulnerability analysis is proposed in [2]. It aims to start cybersecurity investigation in a real hardware manner. The goal of this infrastructure is to ensure secure systems, planning and operation, response, and support. However, in their work, Ussath, M., et al. [3] depict a concept for a security investigation framework to improve the efficiency of investigations of multi-stage Advanced Persistent Threat (APT). The proposed concept utilizes automatic and semi-automatic functions and requires core components such as parsing, visualization, meta information, and correlation components. Also, it collects relevant information from different sources, such as log files, results of forensic investigations, and malware analysis reports. Another study explores the complex investigation of incidents in cloud-based environments [4], in which the automated monitoring tools may implicate rather long lists of virtual machines and containers. The authors propose a visualization approach that aims at reducing the number of VMs and containers awaiting forensic investigation. They conclude that, using the visualization tools, individuals were able to detect malware 70% of the time. Also, future combinations of visualizations with intelligent forensics might provide better results. Rondeau, C., et al. [5] try to improve pre-attack security and post-attack forensic methods in the context of WirelessHART - IIoT (Wireless Highway Addressable Remote Transducer in support of Industrial Internet of Things). They investigate activity aimed at applying Time Domain Distinct Native Attribute (TD-DNA) fingerprinting and improving feature selection to increase computational efficiency and the potential for near-real time operational application. A mini-track overview reports state-of-the-art in the emerging area of cybersecurity investigations and digital forensics [6]. According to the authors, there are still pending challenges in this area. Challenges include the identification of solutions to handle complex investigations, especially those that involve technology such as smart cities, cyber-physical systems, and Internet of Things (IoT) environments. According to Kebande, V., et al. [7] there is a lack of holistic and standardized approaches addressing the digital forensic investigations of IoT systems. In their paper, authors developed and validated a Digital Forensic Readiness (DFR) framework to be used to implement a proactive forensic process in an organization. The presented framework is compatible with the ISO/IEC 27043:2015 standard and considers organizational

132

I. Macedo et al.

complexities and policy developments. On the other hand, a report by Horsman, G. [8] takes a step forward in the field-wide sharing of knowledge in digital forensics. He proposes a framework designed to set out the required elements for sharing reliable digital forensic knowledge, called the Capsule of Digital Evidence (CODE). Due to a study from 2015 by Sandia National Laboratories, USA [9], stakeholders of physical protection systems lack essential knowledge related to cyber-enabled physical attack scenarios. In this study, Sandia R&D team worked to support experiments of cyber-attacks against Physical Protection Systems (PPS) components and subsystems. Their research focuses on developing a reliable capability to investigate cyber-enabled physical attack scenarios: identifying and understanding credible cyber threats to PPS. Furthermore, an emerging sub-discipline of digital forensics was described [10]. This field covers Financial Technologies (or Fintech), and recognize investigations related to financial transactions and payment activities. In another context, there is a rise in cybercrimes related to automated programs, in which, existing tools and research are not capable enough to advance bot crime investigations [11]. In their paper, Rahman, R. U., el al. [11] propose a four-phase web forensic framework to guide forensic examiners to verify that the crime was committed with automated bots. Authors of Adepu, S., et al. [12] introduce a study to investigate the impact of cyber-attacks on a Water Distribution (WADI) system. The developed tool was able to launch multiple attacks and led to a better understanding of attack propagation and behavior of WADI in response to the attacks. Moreover, most investigation research is based on access and system log files, but these logs are often massive, have a flat structure, contain very basic and limited information, and much of their data is not relevant to the investigation [11, 13]. Nevertheless, a relatively old study from 2007 [13] describes an automated attack-tree-based algorithm for eliminating irrelevant information from system log files. The proposed approach also conducts systematic investigations of computer attacks. Besides research proposals and tools, commercial solutions are also available. Gigamon GigaSECURE [14] enables security teams to gain broad access to and control over network data [15]. It uses five security systems to 1) aggregate network traffic, 2) detect the threat and generate events, 3) evaluate events, 4) generate related alerts, 5) capture the resulting process, and 6) investigate the results. On the other hand, Google has introduced a new security investigation tool, in 2018, as part of the Google workspace security center [16]. This tool aims to help admins and security analysts identify, classify, and take actions on their organization’s threats. Beyond that, there is also a lack of theoretical foundations and usage documentation to guide dashboard design, which a study conducted in 2019 tries to tackle. The “What Do We Talk About When We Talk About Dashboards?” report [17] performs a comprehensive analysis of dashboards by characterizing them and the practices surrounding them. This work is essential to support the correct design and implementation of different types of dashboards. Another study from 2018 describes a framework to analyze and develop dashboard templates for enterprises [18]. It focuses on the event/business operation, attribute list, visualization, dashboard capabilities, and the final evaluation of the dashboard.

A Tool to Support the Investigation

133

3 Investigation Approach Our investigation approach regards a specific security workflow and domain. Cyber sensors as well as physical sensors distributed across the system’s infrastructure, collect, daily, vast amounts of events about system activity. To detect malicious behavior and probable attack attempts, event data must be first carefully processed in an efficient manner because it is impossible for a human to analyze such amount of data in an effective way. So, in order to solve this issue, the events originated from the multiple heterogeneous sources of the system are passed to a component designated as Correlation Engine. The Correlation Engine is a pattern matching mechanism that contains expert written rules which are periodically reviewed and updated under a strict protocol. When a set of events trigger a given rule, an alert is originated. Alerts are then sent to an Incident Management Portal (IMP) which provides a user-friendly interface for the SOC Operator. In the IMP, the SOC Operator can visualize generated alerts and query for more related information such as event details. The SOC, with the available information classifies the alerts as either incidents or not triggering an appropriate security response to handle the abnormality. The described systems and their interactions are represented in the UML diagram of Fig. 1.

Fig. 1. Containers view – security ecosystem

The concerned security domain regards three key concepts: Event. Contains low-level information about system activity. Events can have multiple heterogeneous sources, which can be either physical, such as closed-circuit television and access control management systems, or cyber, such as intrusion detection systems and firewall data. Dealing with events from heterogeneous sources is not a straightforward process and requires a lot of investment in terms of data representation and parsing. Alert. Is originated when a set of events trigger a specific security rule. Alerts allow the correlation between events and contain summary information of those same events. They also include a brief justification about how they were generated. Incident. An Incident results of an alert classification by the SOC Operator.

134

I. Macedo et al.

3.1 Architecture Our solution, designated as SMS-I (Security Management Solutions – Investigation) is an investigation tool that acts as an enhanced, fully-automated decision support mechanism. SMS-I gathers information from both alerts and events, producing incident probabilities based on Machine Learning models. It also provides a set of functionalities to the SOC Operator so that he can perform an in-depth inspection about alert occurrences, such as timeline-based dashboards, alert listing, and incident occurrence probability filters. Hence, the SOC Operator when presented with an alert can visualize the probability of it being an incident based on the Machine Learning predictions and can easily perform a thorough inspection of its occurrence, facilitating his work. The SMS-I architecture is described in Fig. 2.

Fig. 2. Containers view – SMS-I internal architecture

The SMS-I comprises several internal modules: Synchronization Mechanism. It is responsible for the data synchronization process, obtaining events and alerts from the Correlation Engine database and incidents from the Incident Management Portal, parsing them into predefined formats, and storing them in the Investigation Database. ML Engine. The ML Engine is responsible for executing Machine Learning models to determine the probability of each Alert being an Incident based both on its own features and on the features of its related events. This information is also stored in the Investigation Database.

A Tool to Support the Investigation

135

Scheduler. The Scheduler orchestrates both Synchronization Mechanism and ML Engine triggering their execution given some constraint (E.g., time-based - every minute). Investigation Database. It contains the data of our system: events, alerts, incidents and Machine Learning results. Kibana. Is the interface of the Investigation Database, it provides several visualization methods that we combined in order to produce intuitive and informative dashboards. Web Interface. It encapsulates the Kibana Dashboards and provides additional functionalities such as interactive listings and filters.

3.2 Machine Learning The ML Engine is a very important component of our architecture since it has the responsibility to execute the Machine Learning models and to store the produced results in the Investigation Database so that they can be accessed by both Web Interface and Kibana. In our work, Machine Learning is used from a supervised perspective and it can be detailed into two different processes, Learning and Predicting. Learning. In the learning phase, events, alerts, and incidents are collected offline and preprocessed into a clean dataset in tabular format. The alerts contain a set of features such as severity, detector and related asset which are enriched with additional features engineered from its related events. The data from the IMP is used to label the alerts accordingly in order to identify which originated incidents and which did not. The produced dataset is then transformed into a comprehensible format so it can be used by the algorithm and split into three sets (train, validation and test) for evaluation purposes. After the training process, the model can use the broad patterns acquired from the data in order to recognize possible incidents on unseen alerts. Finally, the trained model is deployed into the ML Engine component. Predicting. When the ML Engine is executed, it fetches the alerts (and related event data) from the Investigation Database which were not yet inspected by the algorithm. The Machine Learning model computes an incident probability score for each alert which are later stored in the Investigation Database. Several algorithms such as Random Forest, Multi-layer Perceptron and Long ShortTerm Memory were experimented. For a more detailed description of the employed methods readers can be redirected to [22].

4 Visualization Approach A dashboard is a collection of visual displays of information used to quickly monitor conditions and increase general understanding [19]. Visualization dashboards are applied by almost every organization, including airport security centers, financial institutions, and healthcare providers, due to their increasing

136

I. Macedo et al.

importance in a data-driven environment. Their application leads to improved decisionmaking processes, operational efficiency, and data visibility [17]. While designing the dashboards, it is important to consider that 50% of SOC analysts receive less than 20 h of training for year [20]. Thus, there was a conscious effort to keep the visualizations simple and inviting to help the analysts’ performance and their understanding of the investigation tool. 4.1 Dashboards Given their diversity, dashboards can have different purposes, audiences, and visual elements. In this section, the visualization dashboards developed to aid cyber-physical security processes will be presented and justified. Technology Used. Firstly, Elasticsearch is responsible for the analysis, normalization, enrichment and storage of alert and incident data, as well as data provided by machine learning algorithms. All this information is then accessed by Kibana to create new dashboards, which allows the user to search and visualize security data. Both tools belong to Elastic Stack, an open source framework and are easily integrated with one another. The decision to implement these tools is based on their flexibility and ease of installation, configuration, and execution. Kibana represents the graphical interface of Elastic Stack. It applies and combines different visualization methods like graphics, tables, charts, and metrics to build dashboards. Furthermore, the Timelion plugin provided by Kibana allows the analysis of security data in a continuous timeline. Alerts Dashboard. The alerts dashboard includes all data related to security alerts generated by the different threat detection tools available in the SOC. One of this dashboard’s main goals is to monitor the quantity, nature, and severity of alerts, considering their incident prediction probability, which is calculated by the machine learning algorithm. It is important to note that, in this solution, an incident can only have one corresponding alert. However, an alert can be an incident, or it can be irrelevant. According to a report developed by CriticalStart in 2019, more than 70% of analysts feel overwhelmed with the number of alerts and incidents they need to investigate for a day [20]. In comparison, more than 50% of organizations receive over 10,000 alerts daily, which can lead to alert fatigue and neglect [21]. So, to maintain SOC efficiency and reduce the impact of the investigation on the responsible personnel, it is essential to control the number of received alerts. The objective of the graphics and metrics shown in Fig. 1 is to help avoid a sudden overload of alerts by monitoring the total number of cyber and physical alerts. An alert gauge was also developed to ensure that an overwhelming quantity of alerts is not reached. Alert Quantity Monitoring. Figure 3 shows the first set of alert visualizations for the last sixty days. The severity of an alert defines if it should be ignored or if there is a need to carry out a more thorough investigation. Four severity levels were defined, including high, medium, low, and info. Besides controlling the number of alerts for each severity level,

A Tool to Support the Investigation

137

Fig. 3. Alerts dashboard – Alert quantity monitoring visualizations

to avoid the overburdening of SOC analysts, it is also possible to monitor the date of occurrence of alerts. This is useful to perform pattern and trend identification and to study previous incidents and preceding alerts. Alert Severity Monitoring. Figure 4 displays the second important set of visualizations, which monitor the severity of alerts.

Fig. 4. Alerts dashboard – Alert severity monitoring visualizations

The set of graphics and metrics that incorporate Fig. 5 display, from 0% to 100%, the number of alerts that possess a certain probability of being an incident, as well as the average incident prediction probability. In the example shown, most alerts have an incident prediction probability lower than 35%, which leads to a low average probability value. This means that overall, there probably is not any occurrence of an incident. Machine Learning Results. Figure 5 shows the results provided by the machine learning algorithm. This includes the incident prediction probability, in other words, the likelihood of an alert representing an incident. Incidents Dashboard. The incidents dashboard aggregates all detected incidents related to the organization’s security. This dashboard follows the structure of the alerts dashboard by monitoring the quantity, nature, and severity of incidents. However, since

138

I. Macedo et al.

Fig. 5. Alerts dashboard – Incident prediction probability visualizations

there were no incidents detected by the incident handling tools implemented in the SOC, there are no results to display on the dashboard.

4.2 Web Interface To facilitate the security investigation process, we developed a web interface for the investigation tool to extend monitoring, visualization, and evaluation of events taking place within the system. The investigation tool web interface is a dedicated web application that aims to grab all gathered information regarding security concerns and present them through a unified portal. Figure 6 represents the web portal’s main page that contains intuitive links to its available high-level functionalities. Using this web app, security operations users can explore dashboards and information of different levels of details, including logical information, technical specifications, and machine learning forecasting results.

Fig. 6. Web portal – main page

Implementation and Setup. The investigation tool web interface is a standalone web application built using MVC design pattern. It communicates with other internal modules using Restful APIs. The tool is a part of an internal secure network and provides limited

A Tool to Support the Investigation

139

access to a specific type of users, namely SOC operators, being authenticated using another sub-system. Main Features. Following are the main features of the Investigation web application: • Authentication and authorization of logged-in users using a remote authentication module; • Access to interactive dashboards provided by Kibana module, these dashboards are embedded within the investigation tool and include alerts dashboard and incidents dashboard; • Access to a list of alerts of the highest probability to be incidents, provided by the Investigation Database module; • Access to lists of open alerts, closed alerts, and alerts categorized as incidents, provided by the Investigation Database module; • Access to alert/incident specifications, including forecasting results retrieved from the Machine Learning Engine that are also stored in the Investigation Database; • Access and control of user settings and manual trigger of the synchronization mechanism between other modules.

5 Conclusions With this work, we describe how we are using machine learning and the ELK Stack to contribute to the investigation of cyber and physical attacks. A Visualization tool based on Kibana and a web portal, were developed to support Security Operators dealing with the amounts of information coming from both cyber and physical security monitoring solutions. Moreover, we also introduce to the SOC operator machine learning suggestions in order to help him to detect possible incidents. The ongoing work is now the application of this tool to different specific attack scenarios.

References 1. Thakur, K., Qiu, M., Gai, K., Ali, M.L.: An investigation on cyber security threats and security models. In: Proceedings - 2nd IEEE International Conference on Cyber Security and Cloud Computing, CSCloud 2015 - IEEE International Symposium of Smart Cloud, IEEE SSC 2015 307–311 (Institute of Electrical and Electronics Engineers Inc. (2016). https://doi.org/ 10.1109/cscloud.2015.71 2. Nikolov, L.G., Slavyanov, V.O.: Network infrastructure for cybersecurity analysis. Int. Sci. Conf. (2018) 3. Ussath, M., Cheng, F., Meinel, C.: Concept for a security investigation framework. In: 2015 7th International Conference on New Technologies, Mobility and Security - Proceedings of NTMS 2015 Conference and Workshops (2015). https://doi.org/10.1109/ntms.2015.7266478 4. Shropshire, J., Benton, R.: Container and VM visualization for rapid forensic analysis. In: Proceedings of the 53rd Hawaii International Conference on System Sciences (2020). https:// doi.org/10.24251/hicss.2020.783 5. Rondeau, C., Temple, M.: DNA feature selection for discriminating wireless HART IIoT devices. In: Proceedings of the 53rd Hawaii International Conference on System Sciences (2020). https://doi.org/10.24251/hicss.2020.782

140

I. Macedo et al.

6. Glisson, W., Grispos, G., Choo, K.-K.: Cybersecurity investigations and digital forensics: mini-track overview. In: Proceedings of the 53rd Hawaii International Conference on System Sciences (2020). http://hdl.handle.net/10125/64523. Accessed 20 Nov 2020 7. Kebande, V.R., Mudau, P.P., Ikuesan, R.A., Venter, H.S., Choo, K.-K.R.: Holistic digital forensic readiness framework for IoT-enabled organizations. Forensic Sci. Int. Reports (2020). https://doi.org/10.1016/j.fsir.2020.100117 8. Horsman, G.: Part 2:- quality assurance mechanisms for digital forensic investigations: knowledge sharing and the Capsule of Digital Evidence (CODE). Forensic Sci. Int. Rep. (2020). https://doi.org/10.1016/j.fsir.2019.100035 9. Clem, J., Atkins, W., Urias, V.: Investigation of cyber-enabled physical attack scenarios. SAND2015-4202C (Sandia National Laboratories, Albuquerque, New Mexico, USA) (2015). https://www.osti.gov/servlets/purl/1255768. Accessed 24 Nov 2020 10. Nikkel, B.: Fintech forensics: Criminal investigation and digital evidence in financial technologies. Forensic Sci. Int. Digit. Investig. (2020). https://doi.org/10.1016/j.fsidi.2020. 200908 11. Rahman, R.U., Tomar, D.S.: A new web forensic framework for bot crime investigation. Forensic Sci. Int. Digit. Investig. (2020). https://doi.org/10.1016/j.fsidi.2020.300943 12. Adepu, S., Palleti, V.R., Mishra, G., Mathur, A.: Investigation of cyber attacks on a water distribution system. arXiv (2019) 13. Poolsapassit, N., Ray, I.: Investigating computer attacks using attack trees. IFIP Int. Federat. Inf. Process. (2007). https://doi.org/10.1007/978-0-387-73742-3_23 14. GigaSECURE Homepage. https://www.gigamon.com/solutions/gigasecure-security-del ivery-platform.html. Accessed 20 Nov 2020 15. Hamm, J., Liu, A., Ohlhausen, B., Rosenthol, J.: SOC Automation of Threat Investigation [White Paper], Gigamon & Bluvector. https://www.gigamon.com/content/dam/resource-lib rary/english/white-paper/wp-bluvector-soc-automation-of-threat-investigation.pdf 16. Google Workspace Homepage. https://workspace.google.com. Accessed 20 Nov 2020 17. Sarikaya, A., Correll, M., Bartram, L., Tory, M., Fisher, D.: What do we talk about when we talk about dashboards? IEEE Trans. Vis. Comput. Graph. 25(1), 2019 (2019) 18. Noonpakdee, W., Khunkornsiri, T., Phothichai, A., Danaisawat, K.: A framework for analyzing and developing dashboard templates for small and medium enterprises. In: 2018 5th International Conference Industrial Engineering Application ICIEA 2018 (2018) 19. Few, S.: There’s Nothing Mere About Semantics, Perceptual Edge (2017). https://www.per ceptualedge.com/blog/?p=2793. Accessed 18 Nov 2020 20. Daniel, H.: The Impact of Securiy Alert Overload (2019) 21. Casey, T.: Survey: 27% of IT professionals receive more than 1 million security alerts daily. Imperva (2018). https://www.imperva.com/blog/27-percent-of-it-professionalsreceive-more-than-1-million-security-alerts-daily/. Accessed 18 Nov 2020 22. Oliveira, N., Praça, I., Maia, E., Sousa, O.: Intelligent cyber attack detection and classification for network-based intrusion detection systems. Appl. Sci. 11, 1674 (2021). https://doi.org/10. 3390/app11041674

Use of Virtual Reality as Learning Object of Electrical Substations Alexandre Carvalho Silva1 , Arnaldo J. P. Rosentino Jr.2(B) , Alexandre Cardoso1 , and Edgard Afonso Lamounier Jr.1 1 Faculty of Electrical Engineering, Federal University of Uberlândia, Uberlândia, MG, Brazil

{alexandre,lamounier}@ufu.br 2 Electrical Engineering Department, Institute of Technological and Exact Sciences,

Federal University of Triângulo Mineiro, Uberaba, MG, Brazil [email protected]

Abstract. Power substations play a key role in electrical systems. They aim to ensure system reliability, electrical operation without interruptions and safety and protection for the users and system. In this sense, the knowledge of the main concepts associated with substations is essential for the training of students and engineers who work in this area. However, the development of substation courses and training by conventional methodologies does not allow an optimal integration with users. In general, technical visits are used as an auxiliary activity. However, it takes time, since it is usually far from urban centers, where educational institutions are located, and additionally, for safety reasons, it is not allowed being close to the equipment, especially in high voltage substations. In this context, the use of virtual reality may offer educational mechanisms for users to have improved knowledge retention and motivation. Thus, this paper presents a virtual substation as a teaching tool to complement substation courses and trainings. The present system mitigates the impact on the courses and training quality, when it is not possible or it’s difficult the access on site. For example, the scenarios caused by COVID-19, which imposes remote trainings and classes at companies and universities. Keywords: Electrical substations · Learning environment · Virtual reality

1 Introduction The electric substations are responsible for linking the different sections of power system: generation, transmission and distribution. They offer reliability, safety and an adequate power quality to the consumers [1]. A substation comprises transformers, circuit breakers, lightning arresters, instrument transformers, disconnectors, electrical buses, and in some cases reactors, capacitors and FACTS devices. Since this facility is a fundamental section of the electrical system, it is very important to know the main definitions of it. In order to obtain the best performance of students, the institutions must offer an efficient course to guide them properly. However, the most substations course are theorical and it is very difficult for the students have a practical experience with the electrical © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 141–150, 2021. https://doi.org/10.1007/978-3-030-72654-6_14

142

A. C. Silva et al.

equipment and with the substation as a whole. According to [2], practical work on actual equipment tends not to be feasible because of factors such as the following: • • • • • •

Safety of students and professors at electrical substations; Logistics of transporting students; Dependence on the electric utility’s willingness to host technical visits; Schedule conflicts between students, professors, and the electric utility to be visited; The length and frequency of visits being insufficient to meet student needs; High cost of installing and using real equipment in the laboratory.

Besides these issues, even when a technical visit is possible, the knowledge achieved of the concepts that integrates the substation is superficial during the visit. Virtual reality (VR) is a scientific method and technique that human beings create in the exploration of natural processes to understand, simulate, and better adapt and use natural science [3]. So, VR can be used to simulate interactive environments with the advantage of being safe, controlled, and considerably less costly than the purchase of equipment [2, 4]. Following this scenario, this paper presents a learning object for users through a virtual substation, from which is possible: • Identification of the type, voltage class, and function of substation; • Concept and identification of substation bays and its arrangement. In general, the bus arrangement is identified only by single-line diagram; • Identification of equipment inside the substation, showing their operation state. The environment will able the user walking along the substation, simulating a technical visit. Besides, some technical specifications of the equipment can be accessed. With the virtual substation, the users can connect the theory of electrical substations to the practical aspects. In this way, in Sect. 2 are presented some references related to the work. Section 3 presents the applied methodology for implementing the virtual electrical substations. Then, in Sect. 4 is presented the 3D environment of virtual substations. Conclusions of this study are presented in Sect. 5.

2 Related Work Over the years, some works has presented interesting innovations by using VR as training and learning proposal. So, this section aims to present a brief summary of such works and highlight the main contributions of this paper over them. Among the works developed for substations training, a virtual substation with all its equipment was in development to give electricians the opportunity to safely operate the equipment and execute basic and emergence maneuvers in [1]. However, it is a work in progress project, and have a few valuable results. In [2], a virtual reality authoring tool was developed to provide students with a visualization of an autotransformer installed in an electric power substation, and to allow them to perform virtual maintenance and operating procedures on it. So, other equipment are not implemented. In [3] a virtual

Use of Virtual Reality as Learning Object of Electrical Substations

143

simulation system of 500 kV is proposed. The system implements the parameter display of the substation equipment. A non-immersive VR system for training in electrical tests to distribution substations is presented in [5]. It illustrates 40 electrical tests in threedimensional scenes with animations, environmental sound and explanations in text and audio. The system also allows 3D visualization and manipulation of equipment and testing tools. In [6] an Immersive Virtual Environment using HMDs, joysticks or even an ordinary keyboard, mouse and monitor is proposed to provide a realistic experience to trainees in a safe environment where they can interact with equipment, explore the facility and, mainly, practice basic and complex maneuvers to recover substation operations. A real model of a substation of 138 kV in three dimensions using software Sketchup to aid in the learning of students and professionals is proposed in [7]. However, it allows only 3D viewing. In [8] proposes the use of an interactive catalog of elements of an electrical substations, through the use of augmented reality (AR) techniques and access to even through a website, supported by VR. With use of these resources is intended to provide learners in virtual navigation conditions substations, identification of components and its buses arrangements in such environments and access to interactive content related to component data. Based on AR technology and supplemented by virtual reality scenes, the paper on [9] made in-depth research and development of such interactive technologies of wearable devices as data glove, monocular AR glasses, etc. and built a more realistic immersive virtual environment of training for the operation and maintenance personnel of transformer substation from visual sense, tactile sensation, somatosensory sense and interactive mode.

3 Methodology The methodology for implementing the virtual substation is composed of the following stages: • A standardized form, or protocol for the acquisition of information regarding the features of the substation (CAD plans, photos, videos and equipment catalogues); • Norms for the modelling of three-dimensional components aimed at the high performance of systems; • Process for the automatic generation of the three-dimensional environment (automatic positioning of equipment, reuse of setups, topology for equipment start-up); • Standardized interface templates for navigation control, reading of electric component information and command sending. 3.1 Acquisition Protocol One of the main causes of delay in the execution of projects that require 3D modelling is the lack of the necessary documentation for carrying out the process. Among the elements that make up this documentation, one notes: CAD plant design, photos and videos with a field survey of the substations and of the equipment from which it is composed, besides technical equipment catalogues, such as circuit breakers, disconnecting switches, etc.

144

A. C. Silva et al.

CAD Projects. Substations are constructed by means of elaborated projects, which follow technical design norms and protocols that are specific to projects. These however, are documented in digital form, composed basically of two types of files: image files in TIF format (old projects) and CAD files (Computer-Aided Design) for more recent projects, considering 2D floor plans and planes that present elevations along with blank wall and busbar features. These documents, when available, are of great use as they present constructive parameters, which present essential features to 3D modelling: quotas (scale, design), proportions, plan view and side view, planes. By means of CAD documentation, it is possible to exploit vectorial and geographical features of the project. Photos. To achieve a realistic modelling, photographic files are of the utmost importance. Therefore, in order to create 3D models from 2D photos, the minimum of 4 photos from each object is necessary, which one intends to digitalize. These being: Perspective; Right or left; Top; Front or rear and base. As in technical design, an object is broken down into views. On the other hand, it can be achieved through photographs. This technique is known as 3D reconstruction from orthographic views. Videos. The filming of the substation complements the work and provides the modeler greater understanding of the integration of all the equipment as well as its position. For filming, it is recommended that one uses a high definition camera, such as GoPro, with a fisheye lens as it allows for a wider viewing and capture angle, thus obtaining greater detail while permitting that the individual operating the camera narrate what is being filmed, such as the name of the equipment and provide a detailed description, which may be essential to the modeler’s comprehension. Catalogues. Catalogues are documents that are very rich in technical details and of great value to the construction of the virtual model. Therefore, whenever possible it should be used in manner that respects its brand, model and version.

3.2 2D Control Interface for the 3D Environment In order that the user is able to monitor and control the equipment that make up the substation, it is necessary to elaborate two-dimensional control interfaces for performing these activities. In this context, a control interface and selection (menu, Windows, panels and icons) template was elaborated, which attends to every demanded task that is pertinent to the substation management. Therefore, after the generation of the virtual substation scenario, the template containing the full interface description, necessary for performing the activities, is added to it. Through the means of an internal architecture, data referring to the state of the electrical equipment (turned on or off, electrical measurements) are received and processed in real time via WebService. These data are made available by a SCADA system of the Electrical Utility Control Center. Therefore, with this information at hand, the virtual environment is updated with a true representation of the state of the equipment and the control components content.

Use of Virtual Reality as Learning Object of Electrical Substations

145

To prepare this control interface template, a number of requirements were defined, related to usability, layout and design aspects. Those being: • The two-dimensional control interface should be integrated into the three-dimensional virtual environment. • The control interface should promote mechanisms that allow for their quick response use. • The control mechanisms should be presented only when necessary in the context of interaction or when activated by the user. • Production of alternative interfaces for control that possess mechanisms that allow the user activation and deactivation functionality, along with the option of relocation to any desired space within the virtual environment. 3.3 Flowchart for Construction Process A flow process referring to the elaborated methodology is shown by Fig. 1. From the photos and construction documents (CAD, component files, manufacturer documentation etc.), one initiates the construction of 3D models that will go on to make the environments (physical modelling). Each fabricated model is validated and conferred for its insertion into a Library of Models, which further groups together, photos and other documents. A first software was developed for converting CAD models based on virtual environments (semi-automatic generation for virtual environments, by means of the 2D CAD project). Through use of such mechanism, an incomplete VR environment is generated, without cables and connections between the distinct virtual objects. In order to insert the connections, a second software was created, in such a manner as to bring into play the connector cables and automatically connect components, through the topology of the electrical connections. At the end of this process, the virtual environment resembles the actual arrangement in the field (particular to each substation) and is valid for users. Once corrected, the design is sent forward to the association stage, here each element from the virtual model is associated to an identifier within it on the SCADA system, thus generating conditions in the virtual environment, for presenting the state of each monitored component. Finally, through consideration of the authoritative policies, permission of the active control is associated by means of the Interface Template. In situations where there is no Internet access, it is possible to perform simulated operation with historical data. For this situation, data representative of specific situations for simulation were collected. The viewing and navigation mechanisms are independent of online access.

146

A. C. Silva et al.

Fig. 1. Process for the generation of substations

4 3D Environment This section presents the main components of the control interface template, which integrates the virtual electrical substation. 4.1 Selection Bar – Menu and Action Control Panels The elaborated strategy contains a single bar for the selection of control options (menu) situated on the left-hand side of the environment. Fig. 2 presents the item with an active selection bar. Each item that makes up the side bar possesses a panel with options relating to certain actions. In order to visualize this panel, it is necessary only to place the mouse over the opening icon ( ). Thus, it appears only when the action involved is being requested. The button’s features and activation mechanisms for the alternative interface are presented in Fig. 3.

Use of Virtual Reality as Learning Object of Electrical Substations

147

Fig. 2. Active selection bar

Fig. 3. Button and activation mechanism features for alternative interface.

4.2 Control and Content Presentation Windows When the user needs to manipulate a virtual component, windows and control elements are presented after their selection. These windows are also integrated into the virtual environment and their exhibition is finalized automatically after the execution or cancellation of the action, thus creating a less overloaded and more intuitive virtual environment. Fig. 4, Fig. 5, Fig. 6 and Fig. 7 presents such windows.

Fig. 4. Equipment information window

148

A. C. Silva et al.

Fig. 5. Equipment status control window

Fig. 6. Substation circuit breaker closed with status control window

4.3 Interaction with Substation Equipment From the virtual substation, it is possible to provide a new approach for controlling and operating power substation devices, by means of Virtual Reality techniques, which offer greater immersion and more intuitive interactions. Another associated aspect is that the users can navigate in the most diverse ways, exploring and viewing the conditions of the electric components that control the substation with greater safety. Figure 8 Presents the system during the consulting operation of the state of a power transformer.

Use of Virtual Reality as Learning Object of Electrical Substations

149

Fig. 7. Quick equipment locating mechanism and reading information

Fig. 8. Consultation of information concerning the state of a power transformer.

5 Conclusions The methodology described herein was applied successfully to the production of Virtual Reality environments for power utility substations. The system was presented to its main target audience, that is, the system operators. The photorealism observed by the operators was classified as a great level or realism in relation to the field, where the respective substation can be found. In this context, the product has been also used in higher institutions, where is offered the substation course. The achieved results has been

150

A. C. Silva et al.

excellent, especially taking into account the current world scenario caused by COVID-19, where students have been move for remote classes. Acknowledgment. The authors acknowledge the financial support granted by the CAPES, FAPEMIG and P&D ANEEL - CEMIG GT 0618.

References 1. Tanaka, E.H., Paludo, J.A., Cordeiro, C.S., Domingues, L.R., Gadbem, E.V., Euflausino, A.: Using immersive virtual reality for electrical substation training. In: International Conference e-Learning 2015, pp. 136–140 (2015) 2. Barata, P.N.A., Filho, M.R., Nunes, M.V.A.: Consolidating learning in power systems: virtual reality applied to the study of the operation of electric power transformers. IEEE Trans. Educ. 58(4), 255–261 (2015) 3. Liao, X., Niu, J., Wang, H., Du, B.: Research on virtual reality simulation training system of substation. In: 2017 International Conference on Virtual Reality and Visualization (ICVRV), pp. 413–414 (2017) 4. Zhang, X., Jiang, T., Zhang, Z., Liu, C.: Research on Virtual simulation of whole process teaching of thermal power project. In: 2017 International Conference on Smart City and Systems Engineering (ICSCSE), pp. 32–35 (2017) 5. Hernández, Y., Ramírez, M.P.: Virtual reality systems for training improvement in electrical distribution substations. In: 2016 IEEE 16th International Conference on Advanced Learning Technologies (ICALT), pp. 75–76 (2016) 6. Tanaka, E.H., et al.: Immersive virtual training for substation electricians. In: 2017 IEEE Virtual Reality (VR), pp. 451–452 (2017) 7. Lima, G.R., Borges, T.T., Baptista, J.M., Monteiro, P.R.D., Lima, L.: Virtual reality as a tool for electrical substations teaching. In: 2018 Simposio Brasileiro de Sistemas Eletricos (SBSE), pp. 1–6 (2018) 8. Cardoso, A., Júnior, E.A.L., Barcelos, M.A.: Tools to support the teaching of substation components Electricity use with virtual and augmented reality. Rev. Iniciação Científica, Tecnológica e Artística 4(3), 1–8 (2014) 9. Peng, Y., Ding, L., Xu, Z., Jiang, Y., Chen, J.: Design and realization of augmented reality based operation training system for operation and maintenance personnel of intelligent transformer substation. In: 2017 IEEE 2nd Information Technology, Networking, Electronic and Automation Control Conference (ITNEC), pp. 1706–1709 (2017)

Stewart Platform Modified into a Bio-inspirational Device of the Human Shoulder Marius Leonard Olar , Arun Fabian Panaite , Monica Leba(B) and Remus Sibis, anu

,

University of Petrosani, Universitatii 20, Petros, ani, Romania [email protected]

Abstract. This research is geared towards shaping, simulating, designing and developing a movement system, which is based on the principles of the Stewart platform. From the point of view of mobility, it is considered a modified Stewart platform so that as mobility increases, the positions of the static and movable points of the platform change until a form compatible with the human shoulder is reached. Using modelling and simulation, the operation of the modified platform and its compatibility with the human shoulder are justified. To make the real model, there have been used technically and economically accessible materials, to obtain a low-cost product. Unlike the classic Stewart platform, the approach covered by this paper is to simplify, by horizontal optimization, the motion system so that it can synthesize the movement of the shoulder joint, with degrees of freedom similar to the human shoulder, while simultaneously maintaining a reliable force level. The main purpose of this work is to reproduce, first by design, then by calculation, a physical model of the biological mechanism of the shoulder. The model designed in this research replaces the muscles with wires and actuators. In this process, the selection and arrangement of components was simplified in positioning and as number, favoring anatomical mimicry. There was studied and constructed the mechanism according to the anatomical descriptions, which were described from a medical point of view, and there can be noticed that by applying technical mechanics, results the same kind of movements, but not necessarily the same type of joint. Keywords: Stewart platform · Human shoulder · Bio-mimetics

1 Introduction The motivation behind the construction of the Stewart-plated bio-inspirational device stems from the need to have a shoulder joint with realistic biological features and multiple degrees of freedom, in order to manipulate objects in the most natural way. In this project there were used the Robotis AX-12A actuators, because they are optimized for such applications, they present good precision and torque to lift from a horizontal position a robot or human arm, and through repetitive movements, with © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 151–160, 2021. https://doi.org/10.1007/978-3-030-72654-6_15

152

M. L. Olar et al.

mechanical and software optimizations, can improve control methods in solving human motor coordination problems. The development of humanoid robots has renewed interest in anthropomorphic structures, which were until recently neglected by industrial robotics. This is due, in particular, to the complexity of the shoulder, which is the most mobile joint in the human body. Consequently, it increases the role of the shoulder, which is a spherical type joint, directing the arm in abduction-adduction, flexion-extension and internal-external rotation movements, adopting the Stewart platform model, which makes the transition from the abstract model to the natural anatomic one. It is an important objective, in this research, to take over and use the advantages of the natural model, in terms of the functional biomechanics of the anatomical shoulder, which is abstractly reproduced.

2 State of the Art The human shoulder is constrained by a number of muscles and ligaments, which surround the sphere of the joint, and is driven by the balance of tensions and forces. It was thought that the constructive shape of the shoulder could achieve movement with three degrees of freedom with a compact and lightweight robotic arm, which was composed of the usual joints. But many researchers have come to realize that the mechanism did not provide a natural mimicry [1]. The arm designed in this paper was examined repeatedly to evaluate movements for the purpose of determining a natural mimicry. In 1899 a physical shoulder model was built from artificial materials by Mollier and was called the “Shoulder Organ”. This pattern was manipulated from a keyboard, as if “singing” to it. In this old model, the keys were connected to the shoulder joint, with wires corresponding to the muscles that contracted [2]. For the shoulder joint, a lot of research has been done to apply reconstructive surgery and design specifications in many ways. There are also humanoid robots that have been successful in such environments, and whose shoulder joint movements can be compared to those of the human shoulder [3]. For an anatomical arm, it can organize kinematics and dynamics in different ways for different tasks, and also address complex kinematics even in dynamic and unstructured environments [3]. De Sapio et al. provide an extensive analysis of the impact of motion of the scapular belt, associated with the glenohumeral one, both in terms of the requirements, on the kinematics directed by muscle contractions and the momentum of the arms to and from the glenohumeral joint [4]. The shoulder from the paper [5] is a structure that replaced the ball joint with a Stewart platform, optimized to concretely mimic the joint between Scapula and Humerus, and the ligament muscles of that joint. The balance of forces brings movements with 3 degrees of freedom [2]. In contrast to industrial robots, a cable-operated joint is capable of operating a wide set of movements and contributing to the precise handling of objects in dynamic and unstructured environments [6]. Such a motor-operated joint, controlled by cables or wires may be adapted for mounting in an exoskeleton [7] or become part of a robot arm [8]. Some studies have approached conventional prostheses that were attached to the humerus with sockets, compressing the stump, being moved by native muscles without somatosensory feedback [9]. Other applications for the Stewart platforms introduce a

Stewart Platform Modified into a Bio-inspirational Device

153

new controller, aimed at industrial fields, to improve the performance and accuracy of the positioning systems. An example is a Stewart platform used as a mobile platform in non-destructive inspection applications that involves exploring the specimen surface of aeronautical coupons at different heights [10]. The user coupled base and the platform coupled to its proximity, the plurality of structures that extend between the base and the platform are actuated in a parallel order to be moved - resulting an exoskeleton [11]. The rehabilitative robotic exoskeletons, primarily used to consist of a two-link, twojoint device, corresponding to the upper and lower arm and to the shoulder and elbow joints of the human body [12]. Some mechanisms consist of two sliders to allow passive movements of shoulder joints, demonstrating that certain sliders enable shoulder joint to move in the frontal plane with a quarter circular arc during abduction–adduction, while other sliders allow movement of shoulder protraction–retraction as needed. Such mechanisms have been designed to be used, eventually in a 7 degrees of freedom modular robotic exoskeleton, for upper limb rehabilitation [13]. Some upper body exoskeletons emerge as ergonomic auxiliaries for workplace tasks. These exoskeletons consume less and less power, or even no power [14].

3 Modified Stewart Platform Implemented in Shoulder Joint In Fig. 1 is presented the first plane, the fixed plane (Glenohumeral cavity) (1.) and the second plane, the mobile plane (head of the Humerus) (2.). The socket 1 (a.) represents the insertion points of the tendons of the muscles, on the tuberosities of the humerus head (located in the vicinity of the mobile plane), the socket 2 is the insertion point of the Deltoid muscle, on deltoid tuberosity, on the body of the Humerus (located in the first third from the base to the top) (b.) and the socket 3, an imaginary outlet position, is used only as a reference for the other two, and located in the second third from the base towards the top (c.). The turning point of the mobile plane is in the center of the sphere of the humerus head. Socket a., b., and c. are located outside the mobile plane. The human shoulder joint is composed of fixed, supporting, moving, actuating and motion spaces with anchor and pivot points. All these elements, grouped, can form independent systems, which may be fixed or mobile for each other. If the shoulder joint is divided into mobile systems, two systems can be created, one that produces the first moment of motion, and the second leading to the second moment of motion. For this, a device has been made that produces the first moment of movement of the human shoulder (Fig. 1). The first moment of motion is produced in the Glenohumeral joint, performed by the muscles of the rotator cuff, together with the muscles Latissimus Dorsi (back), Deltoid (up) and Pectoralis (front). The second moment of movement is the one produced by dragging the Scapula onto the thorax, made by the muscles located on the back (Rhomboid minor and major, Trapezius), on the neck (Levator scapulae) and thorax (Serratus anterior) and has as pivot point the joint between Acromioclavicular and Sternoclavicular. In this first experiment we performed only the top-down movement, resulting in a model that uses two wires (tendons) and two motors (muscle groups), one for lifting, and the second for anchoring the plane system (in Fig. 1 only the lifting motor is represented). The built device has a Stewart platform in the modified concept version, in the sense that the two planes have the axis that unites the centers, placed on a horizontal axis, with

154

M. L. Olar et al.

Fig. 1. Device with modified Stewart platform (applied to the human shoulder pattern).

planes of the same size (Fig. 2). In the horizontal position (0) the planes are parallel to each other and placed on vertical axes. The first plane (c) is fixed and is the equivalent of the Glenohumeral cavity of the human shoulder joint. The secondary plane (e) is mobile and is the equivalent of the head of the Humerus, which pivots into the Glenohumeral cavity. The space between the two planes (d) is the thickness of the humerus head, with the rotation point in the center of the secondary plane (e). What sets the secondary plane (e) in motion, in an upward horizontal displacement, is the equivalent of the Deltoid muscle (b, g). The deltoid muscle, in the anatomical variant, is fixed to the Humerus by Deltoid Tuberosity. It is positioned at a certain distance from the turning point of the Humerus head, at 1/3 of the base (to the shoulder) and 2/3 of the tip (towards the elbow). In the experiment there was positioned on “Humerus” (f) the attachment area of the “Deltoid muscle” (g) in three points, thus creating three sockets (1, 2 and 3). The undertaken research consists in the analysis of the movement of the Humerus, in the angle formed by the median of the Humerus pivoting at the point of rotation, when raised, caught in each outlet. It has been observed that the median of the Humerus achieves different angles to the horizontal, depending on the socket from which it is raised. Deltoid muscle (b) contracts muscle fiber (g) by lifting the Humerus (f). Muscle contraction (b) is performed by a Robotis AX-12A actuator, and muscle fiber (g) is a 1 mm diameter cable. The Robotis actuator has a different energy consumption (see Table 1) depending on which outlet the cable (g) is connected to. This consumption starts at a nominal value of 0,05 A, and is between 19–20 A at lift and 17–18 A on descent, at a constant voltage

Stewart Platform Modified into a Bio-inspirational Device

155

Fig. 2. Device with modified Stewart platform, consisting of: a. Engine socket; b. Engines (here only the lift engine); c. Main plan, fixed (Glenohumeral cavity); d. Distance between plans; e. Secondary, mobile plan (head of the Humerus); f. Median of the Humerus (longitudinal median); g. Towing cable (median muscle fiber of the Deltoid muscle); 0. The cable’s point of contact on the secondary plane; 1., 2., 3. The sockets positioned along the humerus median (1. Tuberal head tuberosities, 2. Deltoid tuberosity, 3. Imaginary reference location).

Table 1. Energy consumption, lifting, angles and times of movement of the peak of the humerus median, depending on the outlet used Time of plugging in

Energy value (Amps) Raising the tip (cm) Format angle (degrees)

Horizontal - 0

nominal 0,05

8,0

0,0°

At the first outlet - 1

0,19 with 0,17

26,0

30,0°

At the second outlet - 2 0,18 with 0,17

23,5

26,5°

At the third outlet - 3

23,0

26,0°

0,20 with 0,18

of 11,2 V. The Humerus median is 29 cm long. When the cable is in socket 1, the top of the median rises to 26.0 cm, forming a 30° angle. In socket 2, the top of the median rises to 23.5 cm, at an angle of 26.5°, and in socket 3, to 23 cm and at an angle of 26.0°. Only 4 cm of cable was run on the fuse attached to the engine (b) during the lifting of the humerus median (f).

4 Motivation to Re-iterate Platform The modification of the classic Stewart platform consists in reducing to 4 the connecting elements between the planes, and among the 4 points of attachment of the links on the planes, one is outside the plane (Fig. 2). This is because the axis of the platform is horizontal (d) with the planes on vertical axes, and the location of a connecting point

156

M. L. Olar et al.

outside the secondary plane is intended to facilitate the lifting of the fixed element (f) on the secondary plane (e). The experiment highlighted that raising the median with the help of socket 1 leads to an average energy consumption and a large displacement of the median tip. Low energy consumption and an average displacement of the median peak was observed when using socket 2. In the case of socket 3 the energy consumption was high, with the smallest displacement of the median peak. The anatomical Deltoid muscle is fixed to the Humerus on the Deltoid Tuberosity, in the same location with socket 2. Socket 1, even if it has average energy consumption and maximum efficiency, is difficult to control practically, and in the case of socket 3, which also leads to high energy consumption and minimum efficiency. Socket 2, with the lowest consumption and average efficiency, produces the leverage effect which leads to increased control of the handling of an object to be raised with low energy consumption.

5 Movement of the Human Arm As previously stated, the shoulder joint, in the anatomical version, has the greatest mobility of all the joints of the human body. It presents a pattern of ball-type joints in the cavity, and has 6 degrees of freedom, having two moments of motion, one in which the Humerus pivots into the Glenohumeral cavity, and the second, by the Scapula-Clavicle system, which pivots in the upper head of the Stern. For each moment, certain muscles come into action, some agonic, others antagonistic, some synergistic and redundant with the action of other muscles. For example, the Deltoid muscle (median fiber) raises the arm up to the horizontal, then the Suprascapularis muscle continues to lift the arm up to 15° from the horizontal, ending the first moment of motion. The lifting of the arm is continued by the Trapezius muscle, the Romboid and Elevator muscles, up to 120° from the horizontal. In the first moment of motion, the arm moves in the upper transverse plane (Fig. 3), from the horizontal to the front, from 0 to 60°, and towards the back, up to 15°. Then, moves in the posterior sagittal plane (Fig. 4), from horizontal up, 15°, and down to 60°. Continued movement of the arm in all other directions, towards the upper or lower limit offered by the shoulder’s degrees of freedom, leads to the second moment of motion of the shoulder. If this second moment of motion was achieved in a variant of the Stewart platform, a third plane, which will be the basis of the system, should be included as a foreground, fixed, equivalent to the chest and/or spine, and the second plane will become that of the Scapula-Clavicle system. The third plane will be that of the system of the Glenohumeral cavity-head of the Humerus. Each mobile plane shall be operated by 4 motors, with the axis passing through the plane centers, in position 0, horizontally.

Stewart Platform Modified into a Bio-inspirational Device

157

Fig. 3. Upper transverse view (first moment of motion)

Fig. 4. Posterior sagittal vision (first moment of motion)

6 The Modified Stewart Platform The modified Stewart platform presents one of the connecting points of the secondary plane M1 , placed outside the secondary plane. If in the classic version of the Stewart platform AO1 O2 M presents as a parallelogram, as amended AO1 O2 M1 presents as an irregular trapezoid (Fig. 5). The vector d has known and constant length, and the orientation is given by the angles of movement of the shoulder (α, β, γ ).

158

M. L. Olar et al.

Fig. 5. The mathematical model of the modified Stewart platform

We know that AB = BC = CD = AD = l MN = NO = OP = PM = l both are squares. The following vectors are known − → p1 in the reference system O1 x1 y1 z1 − → p2 in the reference system O2 x2 y2 z2 The reference system O2 x2 y2 z2 is rotated with the angles (α, β, γ ) from the reference system O1 x1 y1 z1 having the rotation matrix: ⎡ ⎤ cβsγ −cβsγ sβ R(α, β, γ ) = ⎣ sαsβcγ + cαsγ −sαsβsγ + cαcγ −sαcβ ⎦ sαsγ − cαsβcγ sαcγ + cαsβsγ cαcβ and shifted by the vector d . There are used the notations: sα, sβ, sγ = sinα, sinβ, sinγ , and cα, cβ, cγ = cosα, cosβ, cosγ . In AO1 O2 M there can be writen: − → → − → c1 = d + p2∗ p1 + − → − → c1 and d are in the system O1 x1 y1 z1 p1 , − − → p2∗ in the system O2 x2 y2 z2

Stewart Platform Modified into a Bio-inspirational Device

159

so − → − → → p1 + − c1 = d + p2∗ R(α, β, γ ) − → − → → c1 = d − − p1 + p2∗ R(α, β, γ ) → → → In the same manner there are determined − c2 , − c3 and − c4 , which represent the controls of the four engines to bring the arm into the imposed (α, β, γ ) orientation.

7 Conclusions Because the natural connection point of the Deltoid muscle with the Humerus is attached to the deltoid Tuberosity, for easier lifting of the arm with minimum energy and maximum efficiency, in this research the position of the elevation point was changed from the secondary plane, namely shifted towards the tip of the arm. The experiment carried out highlighted that the positioning of the anchorage point, according to the natural arm, leads to a reduced power consumption, required by the elevator actuator, which, even if it has an average efficiency in lifting an object, compensates with increased control of the movement. In a further development of the device, the actuators needed to perform back-forth movements and the combinations between up-down and back-forth will also be added.

References 1. Ikemoto, S., Kimoto, Y., Hosoda, K.: Shoulder complex linkage mechanism for humanlike musculoskeletal robot arms. Bioinspiration Biomim. 10(6), 066009 (2015) 2. Sakai, N., Sawae, Y., Murakami, T.: A development of joint mechanism of robot arm based on human shoulder morphology. In: The First IEEE/RAS-EMBS International Conference on Biomedical Robotics and Biomechatronics, 2006, BioRob 2006, pp. 982–987. IEEE (2006) 3. Zhao, T.-J., et al.: Research on the kinematics and dynamics of a 7-DOF arm of humanoid robot. In: 2006 IEEE International Conference on Robotics and Biomimetics, pp. 1553–1558. IEEE (2006) 4. De Sapio, V., Warren, J., Khatib, O.: Predicting reaching postures using a kinematically constrained shoulder model. In: Advances in Robot Kinematics, pp. 209–218. Springer, Dordrecht (2006) 5. Stewart, D.: A platform with six degrees of freedom. Proc. Inst. Mech. Eng. 180(1), 371–386 (1965) 6. Artemiadis, P.K., Katsiaris, P.T., Kyriakopoulos, K.J.: A biomimetic approach to inverse kinematics for a redundant robot arm. Auton. Robots 29(3–4), 293–308 (2010) 7. Risteiu, M., Leba, M., Stoicuta, O., Ionica, A.: Study on ANN based upper limb exoskeleton. Paper presented at the 20th IEEE Mediterranean Electrotechnical Conference, MELECON 2020 - Proceedings, pp. 402–405 (2020). https://doi.org/10.1109/MELECON48756.2020. 9140691 8. Olar, M.L., Leba, M., Rosca, S.: Design and control of a biologically inspired shoulder joint. In: World Conference on Information Systems and Technologies, pp. 765–774. Springer, Cham (2020)

160

M. L. Olar et al.

9. Ortiz-Catalan, M., et al.: Self-contained neuromusculoskeletal arm prostheses. N. Engl. J. Med. 382(18), 1732–1738 (2020) 10. Velasco, J., et al.: Experimental validation of a sliding mode control for a stewart platform used in aerospace inspection applications. Mathematics 8(11), 2051 (2020) 11. Hunt, J., Lee, H.: Four-bar parallel actuated architecture for exoskeleton. U.S. Patent Application No 16/387,152 (2019) 12. Shen, Y., Rosen, J.: EXO-UL upper limb robotic exoskeleton system series: from 1 DOF single-arm to (7+1) DOFs dual-arm. In: Wearable Robotics, pp. 91–103. Academic Press (2020) 13. Islam, M.R., Assad-Uz-Zaman, M., Rahman, M.H.: Design and control of an ergonomic robotic shoulder for wearable exoskeleton robot for rehabilitation. Int. J. Dyn. Control 8(1), 312–325 (2020) 14. Van Engelhoven, L., Kazerooni, H.: Design and intended use of a passive actuation strategy for a shoulder supporting exoskeleton. In: 2019 Wearable Robotics Association Conference (WearRAcon), pp. 7–12. IEEE (2019)

Multimedia Systems and Applications

The Design of Social Coupons Used as a Digital Promotional Tactic Joana Sampaio Correia1(B) and Dora Simões2 1 Department of Communication and Art, University of Aveiro, Aveiro, Portugal

[email protected] 2 Digital Media and Interaction Research Centre,

Higher Institute of Accounting and Administration, University of Aveiro, Aveiro, Portugal [email protected]

Abstract. Digital media is evolving and challenging the implementation of corporate communication strategies which ceaselessly look for new tactics for customer acquisition and retention. In this domain, the present study assesses the importance of design in the digitalization of discount social coupons. It focuses on the analysis of the tactics used to increase attractiveness to customers, through an analysis from the point of view of the consumer and the marketeer responsible. The research follows the Design-based Research methodology that requires consequent redesign the artefact proposal – social coupon. Thus, the research is divided in 4 main phases. Data are collected through the inquiries by online questionnaires and interviews, and in a focus group session. The framework created and the evaluation of nine drawn coupon’s proposal shows that consumers have a strong preference for warm colors, rounded shapes, and non-serif fonts. The findings of this research show that marketeers should clarify their tactics and adapt its digitalization to consumers’ preferences. This study can complement previous studies based on the effectiveness of the digitalization of promotional tactics through social coupons. These ones are useful to the companies to understand how to increase consumer’s interest. Keywords: Digital design · Relational marketing · Branding · Consumer behavior · Emotional motivations · Digital age · Design-based research

1 Introduction With the evolution of digital media, corporate communication strategies have also undergone a major change, ceaselessly looking for new tactics for customer acquisition and retention. Thus, this research focuses on the design of digital social coupons as a tactic to increase attractiveness to customers, through an analysis from the point of view of the consumer and the marketeer responsible. The consumer’s motivations and the influence of design on the use of digital discounts are studied. The paper is organized according to the methodology adopted in the research. In the beginning of the document it is presented a brief state-of-the-art focused on the design of digital promotional tactics approaching the motivational behaviors and the emotional design subjects, as well as the generational © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 163–172, 2021. https://doi.org/10.1007/978-3-030-72654-6_16

164

J. S. Correia and D. Simões

changes during the digital age that boosted the evolution of digital marketing. Then, it is introduced the research work and methods through the different phases of the designbased methodology. The paper proceeds presenting the main outcomes of data analysis that it ends showing a digital social coupons proposal and its validation by potential consumers. Also, a critical discussion of the results is presented, as well as the main conclusions, limitations, and future directions of the study.

2 Design of Discount Coupons in a Digital Age The constant desire to communicate anywhere and at all times, as well as the search for greater convenience, impacted the evolution of online marketing and, consequently, online shopping, through digital platforms [4]. This technological mediation in the purchasing market provides greater competition, boosting brands to adopt new solutions and strategies that would allow them to stand out in public. Within these strategies, online promotions have emerged, including social coupons [9]. Social coupons, as the name implies, are online discount vouchers and are available from a variety of companies through daily offer websites. Social coupons allow consumers to enjoy discounts on certain products and/or brands previously selected. According to Niedenthal [12], digital games share certain forms, aims, content and design practices with other media and art forms, which allows for comparison and generalization. The same author adds that similarly to the design of graphical games interfaces, there are many variables that may affect the user’s attraction and performance/usage. Thus, the presentation of the information on the interface should be carefully considered and do not overwhelm the user [8]. So, emotional design in digital coupons is a fundamental concept, given the fact that the quality of these elements can affect the online buyer’s perception and his user experience. Thus, it is important to explore the visual experience on the consumer’s seduction and the connection to emotional reactions that experiences and aesthetic values play [13]. Usefulness and usability are essential, but when deprived of fun and pleasure, of joy and enthusiasm, of restlessness and anger, fear and anger, our existence would be incomplete. Consumers humanize brands in order to build human-like personality traits [11]. FMRI imaging techniques show that a brand’s personality traits may activate areas of the brain associated with emotions, which should be implicated if consumers are thinking about brands and relating them in human-like terms. Thus, the personality of a brand may be represented by consumers as a reflection of activated cognitive categories, when they are exposed to the brand [11]. So, brands are humanized through their design elements, for example. Design elements embrace images, colors, shapes, lettering and animations [15]. Therefore, colors can attract and consumers and shape their perceptions, working as a marketing tool. Through the usage of colors, a brand provides a visual identity, creates strong relationships with its target and position itself against its competitors in the market [10]. Given the constant change in consumer habits and choices, brands need to constantly reinvent themselves, as the only way to achieve the necessary relevance for their “survival”. According to Edelman [5], due to the evolution of the internet, consumers have different engagement with brands and the old way of doing business is unsustainable for marketeers. Customers have individual preferences that require improvement of certain

The Design of Social Coupons Used as a Digital Promotional Tactic

165

criteria for categorizing consumers, in order to understand and predict behaviors, making communication strategies and tactics more persuasive. Also, individuals from younger generations are more present on social network platforms, receiving information from numerous sources. This leads this generation to be the best prepared to manage the avalanche of data that professionals are subjected to today [6]. Alvarez [1] adds that generations Z and Y will have their purchasing decisions strongly influenced by their children and, therefore, even if brands seek to awaken children’s attention and respond to their needs, they cannot fail to please and adapt to previous generations. Therefore, they must integrate communication and seek to respond to different consumption profiles, constantly adapting their forms of communication to technological developments. Given the technological evolution and progression to digital, new strategies for digitalizing promotions have emerged. Thus, brands are developing new tactics in order to attract consumers to the digital versions of their stores.

3 Research Work and Methods In this research, the DBR (Design-Based Research) methodology was adopted. This methodology is traditionally characterized by its multiple iterations of design, development and revision over time [14]. The design cycle begins with the awareness of the problem, which is usually large and complex. Then, the model is evaluated and refined in subsequent cycles. The evaluation method of these cycles is iterative, so the evaluation is not considered a final step. The limitation that occurs after development and the consequent evaluation is an extremely valuable step [2]. In this study the research work is organized in four phases. The first one is an in-depth research that is based on the purpose of the investigation, with the aim of deepening the problem. Thus, it consisted in the realization of the literature review, through the realization of an analysis of the main aspects related to design and their influences in consumer’s preferences, focusing essentially on a problem awareness study. The second phase is focused on the creation of tools with the purpose of analyzing the listed problems and then developing the first data collection tools: a questionnaire directed to inquire final consumers and another one directed to marketeer professional. The first questionnaire is developed in order to obtain consumers’ opinions on the influence of aesthetics in digital discounts. Third phase is centered on a quantitative and qualitative analysis of data collection through the consumer’s questionnaire. The results allowed sharpen up the second questionnaire. This one supporting an interview guide directed to some brands that use this digital tactic (social coupons). The fourth and final phase begins upon the data analysis and is concerned to the design and consequent evaluation of the nine digital coupons’ proposal. The proposal is validated in a focus group session with different generational personas. After the discussion of the main associated problems, the proposal is redesigned, and the investigation had its end. Thereby, it proved that iterative design cycles are fundamental to the methodological process of the present research.

166

J. S. Correia and D. Simões

3.1 Questionnaire The first questionnaire measures consumer’s perceptions through the visual experience of the social coupons. This questionnaire is based on the study of Cassia et al. [3], which is an evaluative study of the effectiveness of social coupons for the recognition of a certain brand. Thus, it has its focus on the design of social coupons and on the aspects that are visually appealing to the public. The sample is characterized by 152 respondents: 106 women and 46 men. Most of them belong to Gen Z (81 respondents), while Gen Y and Gen X have similar number of respondents: 31 of Gen Y and 29 of Gen X. There are only 11 respondents from the generation Baby Boomers. The questionnaire explores some aspects related to the design, studied in the literature review, as shown in Fig. 1 presented below. Those aspects comprehend some elements that the design of social coupons includes, such as: emotional associations with colors, lettering (serif, non-serif, alignments) shapes and aesthetics related with these factors. In order to evaluate the aspects of design listed on the framework, the design of some coupons was presented. Also, the outline of the coupon was evaluated through the presentation of different coupons that only differed in shape and thickness. The questionnaire continues with a question focused on the alignment preferences, through the presentation of 3 different coupons that only differed in the alignment of the text (right, center and left). Another question shows different coupons with distinct fonts, with the purpose of testing different types of fonts: thin vs bold, mechanical vs handwritten; serif vs non-serif and simple vs complex. Then, different color palettes are shown, based on the color of brands studied during the literature review and on the colors of the brands in study. The inquired must choose the three colors that appealed the most. The two following questions are focused on the individual and cultural experiences associated with the colors red and black (negativity vs glamour). The emotional design is evaluated through the colors which are more linked with specific emotions. The survey ends with the feeling of the inquiries about the association between several colors and some brands in study. Thus, it is asked which brands they associate with the two colors of the brands in study: red and yellow. 3.2 Interviews Based on the analysis of the consumer’s opinion through the first questionnaire, the second questionnaire was improved. This questionnaire, as previously mentioned, is an interview script directed to brands and has its focus on the transformation and digitalization of the promotional tactics used by some brands. Thus, it investigates the change of consumer’s needs and motivations’, through the evaluation of the marketer’s vision and purpose on the creation of the brand’s tactics and how to use aesthetics to allure the consumer. The sample had 5 participants: 3 from Sonae, through the online questionnaire and one conducted interview over the phone and 2 from McDonald’s through the online questionnaire. This interview script is divided in four sections: “Design”, “Differentiation of opportunities”, “Digitalization of the promotional strategy” and “Advantages and Disadvantages”. The first section studies the specific criteria used by the brand to design their digital coupons and the attention of the brand towards consumer’s feedback on the app

The Design of Social Coupons Used as a Digital Promotional Tactic

167

stores and social media. In the second section, the segmentation of the consumers is analyzed, and it is investigated if there are differentiation through the discounts offered. It is also asked which age groups most use their discounts. The third section study in depth the progression of the strategies from traditional to digital and it investigates if the regular customers continued with the digitalization of the strategies or if there are changes on the age groups. It is also considered if the usage of these coupons promoted by brands has decreased or increased with the modernization from traditional (paper) to its digital version and if there was an increment of the customer acquisition and retention rates’. The last section is focused on the advantages and disadvantages of these promotional tactics (social coupons). 3.3 Focus Group The new proposal of nine coupons was tested on a focus group session by a group of individuals with different ages and of distinct genders. The participants were characterized by their digital age (Baby boomer, Gen X, Gen Y, Gen Z) and gender (female or male). The focus group was performed at the final phase of the research with the duration of about 30 min. Facing the presentation of the 9 different coupons, the attendants scored of social coupons by their preference, following the norms proposed by the researcher: color, font, shape, type of illustration, the content of the coupon, the language used (seriousness, humoristic or satiric) and other characteristics that were considered relevant and worth to mention. After that, a dynamic conversation was promoted between the researcher and the participants to discuss their preferences, in order to investigate which were their favorites and which were the least relevant proposals and why. The discussion was centered on the “product” analyzed according to consumer’s needs, habits, and context of use.

4 Results 4.1 Consumer’s Perspective Figure 1 identifies the results obtained through consumers inquiring (the green shadows), presented as a framework for the design of more assertive digital social coupons. According to the framework, the coupon’s shape should preferably have a smooth and thick line with straight corners without textures. Also, the asymmetries should be directed to the right side. The text of its content should be aligned to the left. The chosen font must be simple, preferably mechanical, and thin, non-serifed, with greater spacing between characters, not handwritten and avoiding great complexity. Concerning the color, users tend to choose the color palettes of the brands: “coupons.com”, “Beruby” and “Prozis”. The choices represent mostly warm tones and light shades of blue and green, complementing with gray gradients. So, the participants prefer warm tones (red, orange, yellow) that give positive emotions and greater attraction and the white color and bluish tones that transmit greater security, comfort, and tranquility. Thus, it would be relevant to use these colors on the design of coupons.

168

J. S. Correia and D. Simões

Fig. 1. Framework to evaluate the design of digital social coupons and the preferences of the consumers inquired.

4.2 Brand’s Perspective The brands participating in the study are Continente (Sonae group) and McDonald’s Portugal. In general, the two brands tend to respect the criteria and graphic principles of the brand, through a very specific design and corporate identity. Both show great attention to the customer’s opinion, through the analysis of internal and external feedback, obtained in the app stores and by carrying out usability tests, working on an agile methodology. Each brand has different strategies according to its target audience and the needs of its customers. Both Continente and McDonald’s Portugal digitized their promotional tools, giving priority to digital discounts. Consequently, its applications for digital discounts had to undergo revisions, in order to obtain highest customer satisfaction. The main advantages of the digitalization of their promotional tactics are increase of customer’s loyalty and better coupon management through the immediate availability of the discounts everywhere and anywhere. The main disadvantages are the different visibility of the coupons, in comparison to traditional paper coupons and the technological dependency.

The Design of Social Coupons Used as a Digital Promotional Tactic

169

4.3 Digital Coupons Proposal and Validation The results collected through the inquires helped in the visual decisions on the design of 9 coupons (see Fig. 2). These coupons were created through the analysis of the data collected through the diverse phases of the investigation and had its consequent reformulation like an iterative cycle, as the DBR methodology requires. The main criteria used on these coupons were: color, shape, content, and lettering. According to consumer’s choices, they prefer warm colors that captures their attention and gives them safety and positive emotions. Therefore, 3 coupons with warm tones (yellow, orange and red) were developed (coupons with numbers: 4, 5 and 8 on Fig. 2). Bluish and greenish tones also give comfort and tranquility. Thus, these colors are represented through coupons 2, 6 and 9. The coupon number 2 is a mixture between warm and cold tones, through a soft and simple gradient. The last three coupons, numbered: 1, 3 and 7 are based on the same criteria as the previous, except that these ones are more creative and have more contrasting colors, humoristic content and abstract illustrations that can have different interpretations.

Fig. 2. Digital social coupons proposal.

In general, regardless of gender and age, the coupon that seemed to attract everyone’s attention was the 4th shown, with the name: “Oranges” (see Fig. 3). This coupon was mentioned 7 times, and one individual classified it as 1st choice. This participant belonged to the male gender and generation X. On the other hand, this coupon was also classified 6 times in the 2nd choice, by individuals of different ages and genders, which shows a huge preference for this choice in all generations. The huge predilection of participants to coupon number 4 was justified by its color, illustration, readability of the content and creativity. One woman from gen Y, described the colors used as “cheerful and attractive, which capture our attention”, and that “the colors match each other and

170

J. S. Correia and D. Simões

motivate us to buy that product”. In addition, the other participants also expressed that the drawing itself caught their attention more, with an appealing drawing. A man from gen Z described his preferences: “Here I liked the drawings. I thought it goes well with the coupon and the item in question” and a woman from the generation Baby Boomer justified that “the characteristic and style make it appealing”. In addition to the color, design and theme, the fact that the content - oranges - is something of frequent purchase (“product I usually buy”, said a man from gen X) and that all the information is very clear and explicit “the bold highlight of the product discounting also helps”, stated a man from gen Z.

Fig. 3. Segmentation of the focus group.

According to most respondents there is a bigger positive emotional association to the red color than to the black color. Most respondents considered the red color more related to romance than to anger and linked black to glamor and not negativity. Also, individuals identified brands for the red color much more easily than for the yellow color.

5 Conclusions and Future Work The set of data collected contributed to obtain different views of the investigation, allowing to analyze different aspects and points of view. The empirical study allowed to understand the impact of coupons on people and how to get their attention and interest. Through the design and consequent testing of the new coupon proposal, a wider approach was obtained regarding the best use and conception of digital social coupons. The focus on emotional design is fundamental and companies should seek to awaken the consumer’s sensations. The Continente brand prioritizes its customer segmentation, presenting a

The Design of Social Coupons Used as a Digital Promotional Tactic

171

greater differentiation of opportunities in its promotional tactics, demonstrating great customization of discounts for each type of customer. Thus, this brand demonstrates different care for different needs, while the McDonald’s brand does not tend to differ the type of discounts it has been promoting, remaining consistent with its traditional paper version. The bigger disadvantages associated with brands can be assured through the existence of fallbacks (e.g. kiosks next to stores where coupons can be printed) and with the creation of notifications through the application. The two brands guarantee that the digitalization of their discount tactics had a very positive impact, resulting in greater customer loyalty, and an increase in brand awareness and positioning. Since the improvement of research methods of neuromarking tools to collect data, neuromarketing research could have an important role in this investigation, like the evaluation of the infrastructure developed for neuromarketing experiment created by others researchers [16]. The evolution of the pandemic of covid-19 made neuromarketing and eye tracking tests impossible, forcing alternative methods to be used, in order to analyze consumer preferences. However, this could be addressed in future research to complement this framework. Undoubtedly, the pandemic has boosted the technological “leap” and the greater frequency of online purchases [7], intensifying the use of digital discounts and so the relevance of this study. Thus, it allowed a greater openness to consumers to collaborate in the study, given their greater experience of use and knowledge about these digital discount tactics. Also, it would be a benefit for future research, the realization of different tests of the coupon proposals for validation with a greater number of consumers and implementing the coupon proposals on e-commerce sites, facing the technology challenges. Acknowledgments. Further research studies on this issue are detailed on a master’s thesis entitled “The influence of design in the usage of social coupons” on the Department of Communication and Art of University of Aveiro.

References 1. Alvarez, S.: Depois das gerações Z e Y, chega a geração Alpha a ditar as regras do consumo do futuro. Público (2017). https://www.publico.pt/2017/05/04/economia/opiniao/depois-dasgeracoes-z-e-y-chega-a-geracao-alpha-a-ditar-as-regras-do-consumo-do-futuro-1770828 2. Carstensen, A.K., Bernhard, J.: Design science research–a powerful tool for improving methods in engineering education research. Eur. J. Eng. Educ. 44(1–2), 85–102 (2019). https:// doi.org/10.1080/03043797.2018.1498459 3. Cassia, F., Magno, F., Ugolini, M.: The perceived effectiveness of social couponing campaigns for hotels in Italy. Int. J. Contemp. Hosp. Manag. 27(7), 1598–1617 (2015). https://doi.org/ 10.1108/IJCHM-02-2014-0090 4. Dias, G.P., Gonçalves, J., Magueta, D., Martins, C., Araújo, J.: Discount coupons dematerialization : a comprehensive literature review. In: 7th International Conference on Information Process and Knowledge Management, pp. 92–98 (2015) 5. Edelman, D.C.: Branding in the digital age, you’re spending your money in all the wrong places. Harv. Bus. Rev. 88, 63–69 (2010) 6. Expresso: O que a geração Z pode aprender com a Y. Expresso (2014). https://expresso.pt/ queroestudarmelhor/o-que-a-geracao-z-pode-aprender-com-a-y=f865849

172

J. S. Correia and D. Simões

7. Iriani, S.S., Andjarwati, A.L.: Analysis of perceived usefulness, perceived ease of use, and perceived risk toward online shopping in the era of Covid-19 pandemic. Syst. Rev. Pharm. 11(12), 313–320 (2020). 8. Jiang, T., Fang, H.: The influence of user interface design on task performance and situation awareness in a 3-player diner’s dilemma game. PLoS ONE 15(3), 1–4 (2020). https://doi.org/ 10.1371/journal.pone.0230387 9. Kumar, V., Rajan, B.: Social coupons as a marketing strategy: a multifaceted perspective. J. Acad. Mark. Sci. 40(1), 120–136 (2012). https://doi.org/10.1007/s11747-011-0283-0 10. Labrecque, L.I., Milne, G.R.: Exciting red and competent blue: the importance of color in marketing. J. Acad. Mark. Sci. 40(5), 711–727 (2012). https://doi.org/10.1007/s11747-0100245-y 11. MacInnis, D.J., Folkes, V.S.: Humanizing brands: When brands seem to be like me, part of me, and in a relationship with me. J. Consum. Psychol. 27(3), 355–374 (2017). https://doi. org/10.1016/j.jcps.2016.12.003 12. Niedenthal, S.: What we talk about when we talk about game aesthetics. Presented at the Digital Games Research Association (DiGRA), London, UK (2009), DiGRA Online Library 13. Norman, D.A.: Emotional Design: Why We Love (or Hate) Everyday Things (04–2005th ed.). Ingram Publisher Services, US (2005) 14. Pool, J., Laubscher, D.: Design-based research: is this a suitable methodology for short-term projects? Educ. Media Int. 53(1), 42–52 (2016). https://doi.org/10.1080/09523987.2016.118 9246 15. Yang, W., Zhang, L., Mattila, A.S.: Luxe for less: how do consumers react to luxury hotel price promotions? The moderating role of consumers’ need for status. Cornell Hosp. Q. 57(1), 82–92 (2016). https://doi.org/10.1177/1938965515580133 16. Vasiljevi´c, T., Bogdanovi´c, Z., Rodi´c, B., Naumovi´c, T., Labus, A.: Designing IoT infrastructure for neuromarketing research. In: Rocha, Á., Adeli, H., Reis, L., Costanzo, S. (eds.) New Knowledge in Information Systems and Technologies, WorldCIST 2019. Advances in Intelligent Systems and Computing, vol. 930. Springer, Cham (2019). https://doi.org/10.1007/9783-030-16181-1_87

A Tracking Topology to Support a Trajectory Monitoring System Directed at Helping Visually Impaired on Paralympic Athletics Luiz Alberto Queiroz Cordovil Júnior1 , Renan Lima Baima1 , Gabriel Tadayoshi Rodrigues Oka1 , Ana Carolina Oliveira Lima1,2 , Moisés Pereira Bastos1,3 , Ana Isabel Martins4 , and Nelson Pacheco Rocha5(B) 1 Innovation Center in Control, Automation and Industrial Robotics, Manaus, Amazonas, Brazil 2 Institute of Telecommunications, Department of Electronic, Telecommunications

and Informatics, University of Aveiro, Aveiro, Portugal [email protected] 3 Control and Automation Department, State University of Amazonas, Manaus, Amazonas, Brazil 4 Department of Electronics, Telecommunications and Informatics, Institute of Electronics and Informatics Engineering of Aveiro, University of Aveiro, Aveiro, Portugal [email protected] 5 Institute of Electronics and Informatics Engineering of Aveiro, Department of Medical Sciences, University of Aveiro, Aveiro, Portugal [email protected]

Abstract. The study reported by this article was based on previous studies about the development of assistive technology for visually impaired athletes and aimed to evaluate a tracking topology that can be applied for the location and orientation of visually impaired on Paralympic Athletics. The paper presents the tracking topology and its integration in a trajectory monitoring system able to provide vibratory stimulus to visually impaired athletes. Moreover, the paper also presents the results of an experiment that was set up to evaluate a trajectory monitoring system prototype by eight visually impaired athletes when performing their training activities. Keywords: Assistive technology · Visually impaired athlete · Trajectory monitoring · Sensing network · Sensory replacement · Paralympic athletics

1 Introduction Within Paralympic Sports, human guides are essential to help the visually im-paired in athletics races by using ropes that are attached between them and the respective athletes. However, the guides indirectly influence the athlete’s performance, mainly if there are substantial differences in the physical capacity of the athletes and their guides. When this occurs, the athletes may be harmed in their performance. As it was observed by the study reported in [1], the visually impaired athletes are not effectively guided by using a © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 173–182, 2021. https://doi.org/10.1007/978-3-030-72654-6_17

174

L. A. Q. C. Júnior et al.

rope tether. Moreover, problems such as affinity, professional treatments, or the absence of the guides for any reason can cause disruptions in training. Therefore, if there are systems aiming to help visually impaired runners on their normal line or trajectory, the guides would no longer be needed [2]. This study is part of a long-term project aiming the development of digital solutions with tactile sensory feedback to help visually impaired athletes. Categorically, this article presents a tracking topology to be used in a trajectory monitoring system composed by a wearable vibrotactile device and a wireless sensor network, aiming to provide visually impaired athletes with information regarding the correctness of their trajectories during athletics races. Moreover, the article also presents the results of an experimental set up that involved eight visually impaired athletes to determine the adequacy of the proposed concept as the basis of a trajectory monitoring system to support training activities. In the following sections, this article introduces the research problem, the methods that were used, the proposed trajectory topology and its integration in a trajectory monitoring system, the experimental evaluation, and a discussion and conclusion of the results.

2 Problem and Foundations The realization of a movement in space is nothing more than the change of a mobile reference relative to a fixed reference, comprising information about location, orientation, and approach. In turn, mobility is the ability to identify the relationship between one’s position and the objects of an environment and then move autonomously, safely, and efficiently. The most common mobility tool for the visually impaired is the white cane, which is the most traditional and is the universal symbol of visual impairment. However, this tool does not provide the necessary information, such as speed, volume, and distance from objects near the respective user [3]. The other tool option that provides a better route for the blind is the guide dog, as the animal is trained to detect and analyse the complexity of situations such as crossing streets, going down or climbing stairs, potential hazards, among others. Considering the tremendous developments of information and communication technologies, during the last decades, other types of mobility tools can be envisaged, since navigation and orientation systems are commonly noticed in the automotive, nautical, and aviation industries. When it comes to accessibility and autonomy for visually impaired people, several researchers consider correction devices using a virtual interface for orientation. For instance, in [4] a virtual interface for indoor environments using accelerometers from Nintendo Wii controllers is introduced. Other approaches are based on the use of virtual reality and audio as auxiliary tools that provide spatial auditory perception [5–7]. The idea is the possibility of visually impaired people build mental models of the environment from systems that provide auditory information about controlled spaces, objects, or people. In this respect, article [8] presents an interface based on virtual reality, and audio containing information about boundaries and obstacles. Such an application al-lowed the visually impaired to explore

A Tracking Topology to Support a Trajectory Monitoring System

175

environments, discovering points of interest, as well as the development of personalized navigation strategies. Moreover, computational vision techniques are also being used. For instance, in [9] the topic is viewed as a computational vision issue comprising pattern recognition techniques, advanced spatial resolution, and accelerometers as sensors to obtain references of orientation from guide systems. In this context, the authors point out five measurements for obtaining data that have information about the relative distance between two points: rotation angle, minimum and maximum frontal distances, and lateral distances (i.e., left and right distance) [9]. With this information, it is possible to establish indicators of relative distance and can support the orientation of visually impaired people. Finally, knowing that humans can feel different frequencies of tactile vibrations, the literature reports the use of a tactile stimulation mechanism to help visually impaired people to decide the direction to be followed [10]. This was the approach followed in the study reported by this article.

3 Methods 3.1 Issues It is public knowledge that visually impaired athletes perform racing activity on the track by using ropes that are attached between them and their guides. Even if the rope allows the execution of sports activity, the guides indirectly influence the athlete’s performance [1]. Moreover, in the study reported by [11], the authors analysed a set of 186 official championship videos (i.e., World Champion-ships, Paralympics Olympic Championships, and national and regional events) related to 100m sprint races, involving 186 pairs of athletes and guides, to identify the incidents associated to the orientation process based on the rope connection between the visually impaired athlete and the respective guide. Considering the results of this analysis, in almost half of the races some incidents affected the dynamics of the athlete-guide pair. In turn, by analysing the incidents that were identified, it was concluded that, when it comes to location, the main issue is to obtain an accurate measure of the position of the athlete and, when it comes to monitoring, the main issue is related to athlete’s trajectory [11]. 3.2 Proposed Solution Within the context of Paralympic Athletics, a digital solution to help visually impaired was considered to provide an accurate measure of their position in the track to give them security and to indicate them the status of the track. The solution should also get information about the athlete’s direction, position, and provide trajectory monitoring. In this respect, it was designed a tracking topology composed by coordinating nodes and anchor nodes, which was integrated in a system aiming to locate the athletes and to track their trajectories. In addition to the tracking topology, this system comprises: i) a wireless sensor network; ii) a computational algorithm for trajectory correction based on motion capture interfaces; iii) a pairing interface for the location network and actuation devices; and iv) a vibrotactile feedback wearable device to aid visually impaired athletes allowing them to develop a controlled and oriented pace in the race.

176

L. A. Q. C. Júnior et al.

3.3 Experimental Evaluation An experimental set up was conducted involving eight visually impaired athletes to evaluate the tracking topology in the context of the athlete’s training activities. The experiment took place in Manaus Olympic Village, Amazonas, Brazil. By using the system prototype the athletes should not leave the limits of the lanes in which they developed their gaits on the 100 m course and the experimental evaluation aimed to determine the degree of success when using a trajectory monitoring system based on the proposed tracking topology.

4 Tracking Topology and Trajectory Monitoring System In general, there are monitoring systems that use a set of information to estimate the coordinates of a point in different environments, whether outdoor or indoor, employing signal processing techniques in a given communication network. A point where its estimated location needs to be inserted in the coverage area of a communication network and this one, in turn, needs to contain referenced information regarding location. The employment of a triangulation technique for the establishment of a location system requires at least three fixed points with a known location (i.e., in the context of a communication network, there should be three computational nodes that receive the signal from an emitting node) that are defined as the origin of the coordinate system. In terms of the types of processing algorithms of location, there is a combination of centralized and distributed models [12]. Concretely, a network with a large number of nodes is subdivided into smaller networks or clusters that each have a central node for processing the location. Within the cluster, the unknown nodes are located together, and the central node of the cluster is responsible for transmitting the information to the central processor of the network that performs the global location process [13]. By properly balancing the number of nodes in each cluster, it is expected to be possible to save energy through a greater balance between the required processing and the required bandwidth across the network without compromising the result of the location. In a direct correlation, unknown nodes are defined as mobile nodes; the central node of the cluster as the coordinating node is responsible for the estimation technique defined by [14] as collaborative multilateration: “[…] for each location iteration an unknown node from three reference nodes, there is a possibility that one or more unknown nodes do not have reference nodes in their neighbour-hoods. For this reason, a distributed type processing algorithm converges for greater fault tolerance and less possibility of bottleneck formation in the central node of the associated cluster”. According to this approach, the communication network consists of communication modules with defined functions named mobile nodes, coordinating nodes, and anchor nodes (Fig. 1). Real-time data is sent to the anchor nodes located in the vicinity of the track, performing the triangulation from the processing of the signal indicating the position emitted by the mobile node (athlete). Such anchor nodes transmit information regarding the position of the mobile node to the coordinating node, whose purpose is to establish communication for control action through an interface or actuator according to a given communication protocol.

A Tracking Topology to Support a Trajectory Monitoring System

177

The logic comprises the reception, processing, and transmission of the signal through the wireless network and consequently the estimation of the coordinates of a mobile node, considering the entire length of the track, that is, the sectors of straight segments, semicircular sectors, and the eight lanes, thus defining each sector of location. As shown in Fig. 1, the visually impaired athlete when running on the athletics track is posed as the mobile node in the tracking network. A signal transmitted through appropriate communication hardware soundly placed in the athlete’s body, is received by the anchor nodes. The processed packet must contain an indication of distance defined in terms of signal strength transmission of the mobile node to all the anchor nodes. In sensor networks like this, the received signal strength indication (RSSI) is used to establish a relationship be-tween the loss of power of the signal transmitted by the athlete when received by the anchor node and the distance between them [15].

Fig. 1. Tracking topology.

In this context, the following propositions were posed: • Each anchor node in a given location section on the athletics track has defined coordinates. • Every mobile node transmits a signal to the anchor nodes based on the RSSI. • Every anchor node when receiving the RSSI from the mobile node processes the received power, indicating the distance value in meters to the coordinating node for each location sector. • Every coordinating node, when receiving the distance value, performs the triangulation using the solution proposed in [15] and sends the control signal for guidance to the mobile node. The resulting tracking topology was applied in a trajectory monitoring system. For the implementation of the respective prototype a Zigbee network was considered. In

178

L. A. Q. C. Júnior et al.

turn, for the orientation device that receives the signal from the network and promotes the vibration that is represented by a linguistic code, it was used a device that was well accepted and presented reasonable results as an orientation tool in indoor environments with barriers [16]. It is composed by a set of vibrational actuators in contact with the athlete’s skin (e.g., with his backs in the suit mode, or with its arms in the bracelet mode) on both left and right body sides, through which is possible to guide the athletes about the respective directions in order to execute their paths with safety and autonomy.

5 Experimental Evaluation The system prototype was object of an experimental evaluation in the context of training activities of visually impaired athletes in order to determine the adequacy of the proposed tracking topology. In this experimental evaluation, the system feedback was provided by bracelets with vibratory stimulus. It should also be noted that a frequency of 250 Hz was initially considered for the vibratory stimulus, as recommended by the scientific literature [10]. However, after the preliminary test of the vibratory device by the athletes when running, they suggested higher vibratory frequencies. Due to safety reasons, the system was used in addition to the traditional guidance method (i.e., a rope tether). When the training activity consisted in sprints, it was evident that the tracking had a significant delay comparing with the current position of the athlete. How-ever, when the training activity did not imply sprints, the experiment showed that the feedback of the system prototype was consistent with the feedback that the guides provided to the athletes through the ropes. Considering, this scenario, Fig. 1, 2, 3, 4, 5, 6, 7, 8 and 9 (round trip in blue colour and back in red colour) present the performance of the eight visually impaired athletes involved in the experience. The training activity was developed considering the 100 m sprint as denoted at the Figs. 2, 3, 4, 5, 6, 7, 8 and 9 bottom. Athletes 1, 2, 3, 4 and 8 developed a straight/satisfactory gait in the course, while athletes 5, 6, and 7 had deviations in the course. In a more detailed analysis, the paths of athlete 2 (Fig. 3) and athlete 3 (Fig. 4) showed satisfactory results regarding the sense of orientation indicating a straight gait. For athlete 1 (Fig. 2) there was a delay in acting in quartiles 2 and 3, both on the outward and the return routes. Moreover, for athlete 4 (Fig. 5) there was a slight deviation in quartiles 2 and 3 in the outbound direction, although the way back was straight. Finally, despite slight deviations in quartiles 1 and 2 in the round trip (Fig. 9), athlete 8 remained within the limits of the lanes. In turn, for athlete 5 (Fig. 6) there was a sharp deviation in quartiles 2 and 3 on the return route without exceeding the limits of the lane, while the going route of athlete 6 (Fig. 7) was a straight gait, but the return route presented deviations in quartile 2 and 3 since the athlete reached the lateral limits of the lane. Finally, athlete 7 (Fig. 8) had sharp deviations in quartiles 2 and 3 both ongoing and back route.

A Tracking Topology to Support a Trajectory Monitoring System

Fig. 2. Athlete 1.

Fig. 3. Athlete 2.

Fig. 4. Athlete 3.

Fig. 5. Athlete 4.

Fig. 6. Athlete 5.

179

180

L. A. Q. C. Júnior et al.

Fig. 7. Athlete 6.

Fig. 8. Athlete 7.

Fig. 9. Athlete 8.

6 Discussion and Conclusion According to the experimental results it is possible to conclude that, in terms of trajectory monitoring, the tracking topology is adequate since it was found a correspondence between the human guide feedback and the feedback provided by the system being evaluated. However, for sprints there is a significant delay, but the causes of this delay are related to the performance of the communications technology being used (i.e., Zigbee) and not to the tracking topology. In this respect, it should be noted that, when in the presence of sprints, there was a non-linearity and inconsistency of the signal power as a function of the displacement, which directly influences the location estimation process. There-fore, it seems that the topological model and related algorithms are adequate and can be used in terms of some of the training activities of visually impaired athletes. Nevertheless, as a future path, it should be considered the implementation of the proposed tracking topology with other communication technology. Considering the performance of the athletes, five of the eight visually impaired athletes participating in the experiment presented satisfactory performance in terms of linear perception, as they remained within the limits of the streak by actuation of the vibratory device. In a focused analysis of the form of communication, deviations in paths are justified, because conflicts of information and temporal failure in data transmission can

A Tracking Topology to Support a Trajectory Monitoring System

181

occur. In any case, this temporal failure in data transmission does not justify the differences between athletes. One hypothesis is that the vibrating device interface presents some limitations in terms of interaction with their users. This justifies further studies on the usability of the proposed system. One of the relevant aspects that should be considered in a possible usability assessment is the frequency of the vibratory stimulus. In fact, the vibration frequency for the tactile stimulus was initially established as 250 Hz, according to literature recommendations [10], but the athletes required higher vibration frequency. This finding leads to the need to determine the appropriate frequency and to question whether the appropriate vibratory frequency for a running visually impaired should be different from the frequency considered when the same individual is standing or walking at low speed. Acknowledgments. This research was supported by the CNPq (National Council for Scientific and Technological Development) and the resources of the CICARI Lab (Control, Automation and Industrial Robotics Center), Manaus, Brazil. The publication was financially supported by National Funds through FCT – Fundação para a Ciência e a Tecnologia, I.P., under the project UI IEETA: UID/CEC/00127/2020.

References 1. Lima, A.C.O., Gandra, S.R., Leal, E.O.L., do Lago Neto, J.C., Junior, L.: Rope tether usability evaluation with visually impaired athletes in Paralympic race tracks. Int. J. Res. Eng. Appl. Sci. 6, 196–209 (2016) 2. Almeida, N.T.: Pinheiro, V: RFID alarm system and trajectory correction in Paralympic athletics races. Procedia Technol. 17, 170–177 (2014) 3. Welsh, R.: Foundations of orientation and mobility, Tech. rep., American Foundation for the Blind (1981) 4. Evett, L., Battersby, S., Ridley, A., Brown, D.: An interface to virtual environments for people who are blind using Wii technology-mental models and navigation. J. Assist. Technol. 3(2), 30–39 (2009) 5. Wilson, J., Walker, B.N., Lindsay, J., Cambias C., Dellaert, F.: Swan: system for wearable audio navigation. In: 11th IEEE International Symposium on Wearable Computers, pp. 91–98. IEEE (2007) 6. Walker, B.N., Lindsay, J.: Development and evaluation of a system for wearable audio navigation. In: Human Factors and Ergonomics Society Annual Meeting, vol. 49, pp. 1607–1610. SAGE Publications, Los Angeles (2005) 7. Picinali, L., Afonso, A., Denis, M., Katz, B.F.: Exploration of architectural spaces by blind people using auditory virtual reality for the construction of spatial knowledge. Int. J. Hum.Comput. Stud. 72(4), 393–407 (2014) 8. Gomez, J.D., Bologna, G., Pun, T.: Spatial awareness and intelligibility for the blind: audiotouch interfaces. In: CHI 2012 Extended Abstracts on Human Factors in Computing Systems, pp. 1529–1534 (2012) 9. Mascetti, S., Picinali, L., Gerino, A., Ahmetovic, D., Bernareggi, C.: Sonification of guidance data during road crossing for people with visual impairments or blindness. Int. J. Hum.Comput. Stud. 85, 16–26 (2016) 10. Sadatoa, N., Okadaa, T., Kubotac, K., Yonekurad, Y.: Tactile discrimination activates the visual cortex of the recently blind naive to Braille: a functional magnetic resonance imaging study in humans. Neurosci. Lett. 359, 49–52 (2004)

182

L. A. Q. C. Júnior et al.

11. Rego, S., et al.: A digital solution to surpass incidents on 100m sprint for Paralympic visually impaired athletes. In: 9th International Conference on Software Development and Technologies for Enhancing Accessibility and Fighting Info-Exclusion (2020) 12. Patwari, N., Ash, J.N., Kyperountas, S., Hero, A.O., Moses, R.L., Correal, N.S.: Locating the nodes: cooperative localization in wireless sensor networks. IEEE Signal Process. Mag. 22(4), 54–69 (2005) 13. Júnior, L.A.Q.C., Bastos, M.P., Lima, A.C.O.: Network hardware review for location system. Sodebras 11, 256–261 (2015) 14. Savvides, A., Han, C., Strivastava, M.B.: Dynamic fine-grained localization in ad-hoc networks of sensors. In: 7th Annual International Conference on Mobile Computing and Networking, pp. 166–179 (2001) 15. Júnior, L.A.Q.C., Lima, A.C.O., Bastos, M.P., Ferreira, R.S., Gandra, S.R., Oka, G.T.R.: A mathematical approach for distance estimation on athletics track based on signal strength. Sodebras 12, 12–16 (2017) 16. Filgueiras, T.S., Lima, A.C.O., Baima, R.L., Oka, G.T.R., Cordovil, L.A.Q., Bastos, M.P.: Vibrotactile sensory substitution on personal navigation: remotely controlled vibrotactile feedback wearable system to aid visually impaired. In: 2016 IEEE International Symposium on Medical Measurements and Applications (MeMeA), pp. 1–5. IEEE (2016)

Physical Movement Helps Learning: Teaching Using Tracking Objects with Depth Camera Andr´e Sales Mendes1(B) , Luis Augusto Silva1 , H´ector S´anchez San Blas1 , Daniel H. de La Iglesia1,2 , Francisco Garc´ıa Encinas1 , alez1 Valderi Reis Quietinho Leithardt3 , and Gabriel Villarubia Gonz´ 1 Expert Systems and Applications Lab, Faculty of Sciences, University of Salamanca, Plaza de los Ca´ıdos s/n, 37002 Salamanca, Spain {andremendes,luisaugustos,hectorsanchezsanblas, danihiglesias,frangaren,gvg}@usal.es 2 Faculty of Computer Science, Universidad Pontificia de Salamanca, 37002 Salamanca, Spain 3 VALORIZA, Research Center for Endogenous Resources Valorization, Instituto Polit´ecnico de Portalegre, 7300-555 Portalegre, Portugal [email protected]

Abstract. Physical activity is the main element affecting students’ cognitive-motor skills. In several studies, it has been shown that physical exercise has a positive influence on health as well as offering an improvement in intellectual performance. In this paper, a new teaching methodology that encourages physical activity is proposed and discussed. To achieve this, the learning process has been gamified by using balls to interact with educative activities. Balls hits on the activities projection are detected using image recognition with a Kinect depth camera. To validate the system, tests have been carried out with a group of school students. As a result, a complete low-cost system that allows teaching educational content while motivating movement has been developed. Keywords: Kinect

1

· Physical activity · Learning · Students · Teaching

Introduction

Currently, sedentarism is one of the main causes of mortality in the population [1]. Also, obesity is one of the main problems of the child population largely caused by a lack of exercise [2]. In addition to metabolic disorders, the main cause of obesity is the combination of excessive caloric intake and decreased energy expenditure over time. To combat and prevent obesity, it is important to look at the underlying causes of excessive intake and decreased physical activity. Previous research has identified several factors, including a severe lack of time, decreased opportunities for physical activity, and sedentary lifestyles [3]. To search for solutions to this problem, several studies have been carried out. Some concluded that moderate and regular physical activity decreases the risk c The Author(s), under exclusive license to Springer Nature Switzerland AG 2021  ´ Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 183–193, 2021. A. https://doi.org/10.1007/978-3-030-72654-6_18

184

A. S. Mendes et al.

of suffering this type of disease [4]. Also, it has been shown that regular physical exercise has a positive influence on learning by improving cognitive processes [5,6]. In recent years, research on TEL (Technology-enhanced learning) has focused on new technologies such as augmented reality, ubiquitous learning (u-learning), mobile learning (m-learning), or serious games [7]. However, despite technological advances and new infrastructure in schools, none of the learning methodologies succeeded in combining moderate physical activity with structured learning related to the subjects taught in the course. There are examples in the field of educational video games and games that combine physical activity with the learning process [8,9]. This article proposes the use of an educational tool based on motion detection through the Kinect sensor on a game that is projected on the wall of the classroom. Students use balls to hit the projected elements. These collisions are detected by the sensor and recorded in the program, thus completing the task at hand. This article is organized as follows: Sect. 2 focuses on an in-depth review of the state of the art on video games in education, sensors, and existing educational systems. Section 3 describes the architecture of the proposed system. Techniques and tools used for the implementation and validation of the system in a real environment are described in the Sect. 4. Finally, the conclusions reached and future works are presented in Sect. 5.

2

State of the Art

The use of video games as an educational resource is a subject studied in multiple works [10,11]. There is a great variety of serious games aimed at different disciplines. Among all of them, we can find those that promote physical exercise, the so-called exergames. The application of exergames in the classroom, as well as their effectiveness, has been widely studied in [12–14]. Among exergames, several technologies have been used, such as the Kinect sensor and the Nintendo Wii console. Corresponding to the latest technology mentioned, we find the case of [15] where they analyze a series of cooperative and competitive games to reduce obesity in teenagers. The appearance of the Kinect sensor by Microsoft caused a great proliferation of serious games. The great benefit of the usage of this type of sensor within classrooms can be seen in the work [16]. In this work, they highlight some of its advantages among which is the interaction with several users at the same time without the need to use a mouse or a keyboard. Also, they study their use to encourage student learning while performing physical exercise. In the current literature, there are also studies focused on analyzing the benefits of Action Video Games (AVG) in the classroom with students. In the paper [17], the authors analyzed the progress of a group of 95 fourth-grade students. The students performed AVG activities supervised by teachers (e.g., Wii Sports or Xbox Just Dance). These students participated in 50-min sessions per week for a total of 6 weeks. Teachers evaluated the students before and after the activity

PHL: Teaching using Tracking Objects with Depth Camera

185

on their academic progress, attention, behavior, and social and emotional performance. As a result, it was observed that children scored significantly higher on post-AVG activity tests than on previous tests. Similarly, the authors of [18] have demonstrated, through the performance of activities with 53 elementary students, that there is an improvement in performance after performing activities with AVG. Furthermore, it has been shown that other parameters such as physical fitness, reaction times, spatial perception, and levels of enjoyment increase with this type of activity [19]. Not only the improvement of academic performance has been demonstrated by different authors, the improvement in the physical condition and its impact on health has also been widely demonstrated thanks to the integration of AVG in the classroom. In the work [20], the authors have noted an improvement in a group of elementary students in terms of flexibility, strength, and grip of the students. After the analysis of more than 260 articles the authors of [21] have concluded that, in general, AVGs showed positive effects in the improvement of physical and body activity among overweight or obese youth. The use of new tools for the promotion of physical exercise such as AVG can also have a positive impact on mitigating some of the effects produced by the COVID-19 pandemic. Due to significant mobility and confinement restrictions worldwide, many people have been unable to engage in outdoor physical activity for months. Staying active at home with aerobic exercise on a stationary bike or treadmill can help counteract the physical and mental side effects of COVID-19 confinement. It has been shown that activities performed with AVG can be a valid alternative to maintaining physical activity from home without the need for specialized fitness equipment [22]. All the activities mentioned throughout this section are mainly focused on making the user interact with the game using his body. However, it is difficult to find applications that analyze the movements and offer feedback on how to improve to the user. By adding this system, it would be possible to analyze the performance of users so that a series of exercises adapted to them could be offered. A good area where to apply this type of application is in the learning of dances or martial arts [23]. Another work is done by [24] where a personal monitor is created to help the correct learning of martial arts. Another environment in which this system can be applied is professional athletes. For example, in [25], authors make use of motion capture technology (MOCAP) to observe the performance of athletes and predict possible injuries.

3

Proposed System and Architecture

This section will present the system and architecture designed to facilitate the work of teachers in promoting physical activity in combination with learning in the classroom. The proposed system is based on projecting educative activities and allowing students or users to interact with them by exercising. The form chosen for interaction with these contents has been through the use of balls. To guarantee a low cost of the system, we decided that balls should not have any

186

A. S. Mendes et al.

kind of internal sensor. This does not only allow the use of balls that are already available in the high-schools but also allows the centers that intend to teach using the system to do so with the least possible cost. The challenge of the system is to detect the interactions and communicate the different parts. To solve this, the proposed architecture is based on the use of image recognition techniques for the detection of these interactions. On the other hand, the system architecture is designed so that the system can be deployed in as many classrooms as possible at the lowest cost by taking advantage of already installed hardware and common computer equipment. The architecture of the system, divided into 4 main modules, can be seen in Fig. 1.

Fig. 1. System architecture

Capture and Representation Services: This module represents the computer equipment that must be deployed in the game area to perform the representation of the contents and the capture of the iterations. The main component of this module is the central computer that can be any traditional computer or a low-cost microcomputer such as Jetson Nano. This device is optimized to perform the processing of images. The necessary components in this module apart from the central computer are the projector and the Kinect. The projector displays the contents on a large wall or surface, where it can be interacted with by using balls without damaging the surface. The Kinect has been chosen because it is a low-cost device and allows the capture of images with a resolution of 1920 × 1080 px at 60 fps and a resolution of the camera depth of 512 × 512 px.

PHL: Teaching using Tracking Objects with Depth Camera

187

The Kinect can be easily replaced by another device with the same characteristics such as the Orbec Astra camera or the RealSense camera from Intel. The main advantage of using this type of device compared to the use of a traditional camera is that, in addition to the color of each point on the image, Kinect captures depth information for each of them. The system communicates with the cloud services to translate these color and depth images into user interactions. Cloud Services: this module represents the services implemented to store and process data. This module consists of a non-relational database, MongoDB. And an API that exposes the web services both for identifying and capturing user interactions to the configuration, monitoring, and exercise planning applications. This module also has a web server that is responsible for serving these applications, and a proxy balancer to distribute requests. User Applications: this module contains the system monitoring application. In this application, the teacher can configure the exercises with their planning and see the results with the success rate of each exercise. Game Area: this module refers to the environment where the sensors and elements used by the students are deployed.

4

Use Case and Preliminary Results

This section describes the techniques and tools used for the implementation and validation of the system in a real environment. For the operation of the system, as mentioned above, there are two main elements in the system: the projection of the contents and the recognition of the user interactions with the system. For the development of the application’s representation and capture system, the .NET Framework environment has been chosen because it can currently export to different platforms such as Windows, Mac, or Linux including ARM systems. This way, the system proposed is compatible with microcomputers such as Jetson Nano. In addition to this advantage, most depth camera manufacturers provide a native SDK for this environment. For the processing of images, Emgu CV has been used. This is an OpenCV wrapper for .NET portable to different platforms and is dual-licensed under both an open-source license and a paid commercial license. User interaction is based on detecting the position of the balls and their collisions with the elements projected. To achieve this, real-time video processing techniques were used. The system allows students to be grouped into teams. So, it is necessary to identify which elements belong to each team or student. This differentiation of the element is achieved by using balls of different colors so that they are easily distinguished. In Fig. 2, the procedure to detect each of the balls and validate that they are colliding on the wall is presented. The system bases its operation on two different elements offered by Kinect: the image frame and the point cloud. The use of the point cloud allows us to know how far from the camera is each of the captured pixels.

188

A. S. Mendes et al.

Fig. 2. Input detection

To calculate the position of the balls, each video frame must be analyzed. The procedure used to analyze them is the following: First, the video frame is scaled so that its size matches the size of the point cloud. This scaling also makes computations using the video frame more affordable to low-end devices. Next, a color filter is applied to each frame for every user or team. Each of these filters uses a different HSV color range to distinguish balls of different colors. Then, a contour finding algorithm is applied to each of the filtered images. After this, the center and the size of each contour is calculated. These would be estimations of the centers and the size of the balls in the image. Later, the depth of each ball is approximated using both the area of the ball (balls farther from the camera have smaller areas than balls closer to the camera) and the point cloud provided by the depth camera. Using this dual-source approach to estimate the depth helps to discard false positives in the system. Finally, the centers of each ball are transformed from the coordinate system of the camera to the coordinate system of the application being projected. Ball collisions are translated into clicks on a web browser embedded in the application. This way, some existing games are already compatible and the complexity of the motion capture system is transparent to the developers of new games. 4.1

Complete System

The complete system developed can be seen in Fig. 3 in the laboratory being tested. The games that are currently in the system have been developed for educational purposes. These games can be seen in Fig. 4. The development of these games has made use of the Phaser JavaScript library.

PHL: Teaching using Tracking Objects with Depth Camera

189

Fig. 3. System in laboratory

– The first game is a game that allows teaching geography concepts in a very interactive way. The user is shown a map with a selected province and some answers. The objective is to select the correct province. – The second game developed is a questionnaire where the content is configurable from the platform. In this game, students have to answer the questions raised with one of the options given. – The third game is the well-known Tic-tac-toe where students are encouraged to coordinate their turns. – The last game is a game where the main objective is to hit the elements that appear on the screen. This game is designed to encourage teamwork. The developed system has been tested in a real environment, can be show in Fig. 5, where the users were children from a school between 9 and 13 years old. Professors from the Education Department of the University of Salamanca have collaborated in these tests. The tests carried out consisted in measuring student motivation when performing the same type of exercises in both its traditional format and using our system. Our tests have shown that students were paying more attention and were more motivated when using our system than using traditional methods. Also, our system was performant handling more than 15 users simultaneously.

190

A. S. Mendes et al.

Fig. 4. Developed games

Fig. 5. System in real environment

5

Conclusions

Analyzing the results obtained, it can be said that the system designed allows promoting physical activity in schools while teaching educational content at the same time. Furthermore, it has been possible to build a system that is affordable for all schools since they can reuse the devices they already have or buy new ones for a relatively cheap price. The use of image recognition techniques in conjunction with the depth of field camera has helped to create a fully functional system. The tests carried out in the educational center have allowed proving that the system impacts education in a positive way since students were more motivated. However, the results of these tests are preliminary. Therefore, more tests should be carried out in order to obtain a greater number of results. This would allow

PHL: Teaching using Tracking Objects with Depth Camera

191

us to know, to a greater extent, the effectiveness of the system as well as to improve it by adapting it to the different situations that may arise. In future works, the system could be integrated with platforms that are being currently used in education, such as Moodle, or improved by adding selfevaluation systems. Also, other projects involving more students and teachers are being considered to expand this research line. In this way, the aim is to broaden the horizon of study in search of results that will improve education in courses with children. Acknowledgment. This work was supported by the Junta De Castilla y Le´ on— Consejer´ıa De Econom´ıa Y Empleo: System for simulation and training in advanced techniques for the occupational risk prevention through the design of hybrid-reality environments with ref. J118. Andr´e Filipe Sales Mendes’s research was co-financed by the European Social Fund and Junta de Castilla y Le´ on (Operational Programme 2014–2020 for Castilla y Le´ on, EDU/556/2019 BOCYL). Francisco Garc´ıa Encina’s research was partly supported by the Spanish Ministry of Education and Vocational Training (FPU Fellowship under Grant FPU19/02455).

References 1. Ng, M., Fleming, T., Robinson, M., Thomson, B., Graetz, N., Margono, C., Mullany, E.C., Biryukov, S., Abbafati, C., Abera, S.F., et al.: Global, regional, and national prevalence of overweight and obesity in children and adults during 1980–2013: a systematic analysis for the global burden of disease study 2013. Lancet 384(9945), 766–781 (2014). https://doi.org/10.1016/S0140-6736(14)60460-8 2. Ebbeling, C.B., Pawlak, D.B., Ludwig, D.S.: Childhood obesity: public-health crisis, common sense cure. Lancet 360(9331), 473–482 (2002). https://doi.org/10. 1016/S0140-6736(02)09678-2 3. Sabin, M.A., Kao, K.T., Juonala, M., Baur, L.A., Wake, M.: Viewpoint article: childhood obesity-looking back over 50 years to begin to look forward. J. Paediatr. Child Health 51(1), 82–86 (2015). https://doi.org/10.1111/jpc.12819 4. Ward, Z.J., Bleich, S.N., Cradock, A.L., Barrett, J.L., Giles, C.M., Flax, C., Long, M.W., Gortmaker, S.L.: Projected us state-level prevalence of adult obesity and severe obesity. N. Engl. J. Med. 381(25), 2440–2450 (2019). https://doi.org/10. 1056/NEJMsa1909301 5. Hillman, C.H., Buck, S.M., Themanson, J.R., Pontifex, M.B., Castelli, D.M.: Aerobic fitness and cognitive development: event-related brain potential and task performance indices of executive control in preadolescent children. Dev. Psychol. 45(1), 114 (2009). https://doi.org/10.1037/a0014437 6. Stevens, T.A., To, Y., Stevenson, S.J., Lochbaum, M.R.: The importance of physical activity and physical education in the prediction of academic achievement. J. Sport Behav. 31(4), 368 (2008) 7. Consortium, N.M., et al.: Horizon report: 2014 higher education edition. austin, tx: New media consortium (2017). http://www.nmc.org/pdf/2013-horizon-reportk12.pdf

192

A. S. Mendes et al.

8. Fernandes, L.M.A., Matos, G.C., Azevedo, D., Nunes, R.R., Paredes, H., Morgado, L., Barbosa, L.F., Martins, P., Fonseca, B., Crist´ ov˜ ao, P., de Carvalho, F., Cardoso, B.: Exploring educational immersive videogames: an empirical study with a 3D multimodal interaction prototype. Behav. Info. Technol. 35(11), 907–918 (2016). https://doi.org/10.1080/0144929X.2016.1232754 9. S´ aez-L´ opez, J.M., Miller, J., V´ azquez-Cano, E., Dom´ınguez-Garrido, M.C.: Exploring application, attitudes and integration of video games: MinecraftEdu in middle school. Educ. Technol. Soc. 18(3), 114–128 (2015). http://www.ifets.info/journals/ 183/9.pdf 10. Amory, A., Naicker, K., Vincent, J., Adams, C.: The use of computer games as an educational tool: identification of appropriate game types and game elements. Br. J. Edu. Technol. 30(4), 311–321 (1999). https://doi.org/10.1111/1467-8535.00121 11. Charsky, D., Ressler, W.: “games are made for fun”: lessons on the effects of concept maps in the classroom use of computer games. Comput. Educ. 56(3), 604–615 (2011). https://doi.org/10.1016/j.compedu.2010.10.001 12. Whitehead, A., Johnston, H., Nixon, N., Welch, J.: Exergame effectiveness: what the numbers can tell us. In: Proceedings of the 5th ACM SIGGRAPH Symposium on Video Games, pp. 55–62 (2010). https://doi.org/10.1145/1836135.1836144 13. Di Tore, S., D’elia, F., Aiello, P., Carlomagno, N., Sibilio, M.: Didactics, movement and technology: new frontiers of the human-machine interaction (2012). https:// doi.org/10.4100/jhse.2012.7.proc1.20 14. Mu˜ noz, J.E., Villada, J.F., Giraldo Trujillo, J.C.: Exergames: a technological tool for the physical activity abstract. Revista M´edica de Risaralda 19(2), 126–130 (2013). https://doi.org/10.22517/25395203.8527 15. Staiano, A.E., Abraham, A.A., Calvert, S.L.: Adolescent exergame play for weight loss and psychosocial improvement: a controlled physical activity intervention. Obesity 21(3), 598–601 (2013). https://doi.org/10.1002/oby.20282 16. Hsu, H.M.J.: The potential of kinect in education. Int. J. Inf. Educ. Technol. 1(5), 365 (2011). https://doi.org/10.7763/IJIET.2011.V1.59 17. Gao, Z., Lee, J.E., Pope, Z., Zhang, D.: Effect of active videogames on underserved children’s classroom behaviors, effort, and fitness. Games Health J. 5(5), 318–324 (2016). https://doi.org/10.1089/g4h.2016.0049 18. Sun, H., Gao, Y.: Impact of an active educational video game on children’s motivation, science knowledge, and physical activity. J. Sport Health Sci. 5(2), 239–245 (2016). https://doi.org/10.1016/j.jshs.2014.12.004 19. Coknaz, D., Mirzeoglu, A.D., Atasoy, H.I., Alkoy, S., Coknaz, H., Goral, K.: A digital movement in the world of inactive children: favourable outcomes of playing active video games in a pilot randomized trial. Eur. J. Pediatr. 178(10), 1567–1576 (2019). https://doi.org/10.1007/s00431-019-03457-x 20. C ¸ ifci, C ¸ ., Ba¸spınar, S.G.: The effects of active video games on strength, vertical jumping and flexibility in children aged 12 to 15 years old 1. Int. J. Appl. Exerc. Physiol. 9(6), 64–71 (2020). https://doi.org/10.1097/00005768-200203000-00015 21. Gao, Z., Zeng, N., McDonough, D.J., Su, X.: A systematic review of active video games on youth’s body composition and physical activity. Int. J. Sports Med. (2020). https://doi.org/10.1055/a-1152-4959 22. Hammami, A., Harrabi, B., Mohr, M., Krustrup, P.: Physical activity and coronavirus disease 2019 (COVID-19): specific recommendations for home-based physical training. Managing Sport Leisure 1–6 (2020). https://doi.org/10.1080/ 23750472.2020.1757494

PHL: Teaching using Tracking Objects with Depth Camera

193

23. Ogawa, T., Kambayashi, Y.: Physical instructional support system using virtual avatars. In: Proceedings of the 2012 International Conference on Advances in Computer-Human Interactions, pp. 262–265 (2012). ISBN 978-1-61208-177-9 24. Chye, C., Nakajima, T.: Game based approach to learn martial arts for beginners. In: 2012 IEEE International Conference on Embedded and Real-Time Computing Systems and Applications, pp. 482–485. IEEE (2012). https://doi.org/10.1109/ RTCSA.2012.37 25. Pueo, B., Jimenez-Olmedo, J.M., et al.: Application of motion capture technology for sport performance analysis (2017). https://doi.org/10.47197/retos.v0i32.56072

Authoring Tool for the Creation of Learning Objects in Occupational Education to Support People with Cognitive Disabilities Paola Pachacama1

, Luis Cuascota1

, and Graciela Guerrero1,2(B)

1 Departamento de Ciencias de la Computación, Universidad de las Fuerzas Armadas ESPE,

Sangolquí, Ecuador {pmpachacama,lscuascota,rgguerrero}@espe.edu.ec 2 Departamento de Sistemas Informáticos, Universidad de Castilla – La Mancha, Albacete, Spain

Abstract. Assistive technology is focused on supporting the restoration or improvement of functionalities of people with disabilities, thus achieving a better quality of life and optimal performance in educational, employment and social activities. This research addresses the issue of employment education for people with cognitive disabilities within Specialized Education Institutions and at home, as well as the limitations that tutors have in managing the content of technological applications because most of them are designed for a specific activity and do not allow the tutor to adapt it in a personalized way based on the needs of people with cognitive disabilities. It was proposed the development and evaluation of an author’s tool called “TK-System”, which allows tutors to manage learning objects through a web application so that they can be used by people with cognitive disabilities through a mobile application. For the execution of usability tests and questionnaires users were considered based on their knowledge (care of people with cognitive disabilities and IT), the results allowed to determine that the tool is useful and easy to use for the creation of learning objects, in addition to obtaining recommendations for improvement. Keywords: Cognitive disabilities · Assistive technology · Learning objects · Author tool · Special education

1 Introduction Currently in Ecuador, the Specialized Education Institutions (IEE) together with relatives of people with cognitive disabilities are in charge of teaching and strengthening competencies, academic, functional and occupational skills related to work. These skills are necessary for the proper development of people with cognitive disabilities in the social and labor fields, all in compliance with the provisions of the Ministry of Education, through a National Model of Management and Care for Students with Special Educational Needs [1]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 194–204, 2021. https://doi.org/10.1007/978-3-030-72654-6_19

Authoring Tool for the Creation of Learning Objects

195

Assistive technology [2] in the education of people with some kind of disability has become a fundamental support tool, since it allows improving the quality of the teachinglearning process by enhancing intellectual and affective skills of people with cognitive disabilities. Tutors are in charge of following up and supporting the difficulties presented by people with cognitive disabilities in the use of these tools, as well as looking for ways to design and adapt learning objects to the students’ needs. However, in the case of IEE and at home, technology has been little involved in the educational process because in many cases the technological tools are designed for only one type of predefined activities, thus avoiding their modification and adaptation for other types of activities according to the needs of people with cognitive disabilities. In the work carried out by Cuascota et al. [3] a technological solution is proposed to support the learning and execution of daily activities of the students of the Fundación Virgen de la Merced (FUVIME, Sangolquí-Ecuador). For this purpose, Beacons devices were used in conjunction with a mobile application called TK-2, which consists of an activity module (origami) and the other module was composed of the presentation of results applying techniques that gamification. The study showed a significant improvement in performance in the execution of activities of people with cognitive disabilities, however, the application is still limited to a single predefined activity and teachers cannot manage the existing content within it. The present work proposes to develop an extension of the work developed by Cuascota et al. [3], in this extension it is proposed the development and evaluation of an author’s tool that allows tutors of people with cognitive disabilities to manage learning objects that can be used through the TK-2 application and has functionalities such as: user management, measurement and reporting of results. This article consists of the following sections. Section 2, Works related to technological tools for the education of people with cognitive disabilities Sect. 3, Description of the implementation and architecture of the tool. Section 4, Design and application of usability tests, Sect. 5, Interpretation of results and Sect. 6, Conclusions and future work.

2 Related Works For information on previous work or work related to this research, a search is made in digital repositories based on the following criteria: people with cognitive disabilities, occupational education for people with cognitive disabilities, assistive technologies applied in special education. For the selection of the most relevant works, the application of inclusion and exclusion criteria is considered; Inclusion: i) Works focused on the support to the teaching of people with cognitive disability ii) Works with technological devices Exclusion: i) Studies not peer-reviewed ii) Studies not completed iii) Studies not proven to be scientifically supported. In the study conducted by Chan et al. [4], they proposed the design, implementation and evaluation of an Alternative Augmentative Communication (AAC) system in conjunction with beacons and IOT devices, to facilitate the communication of children with cognitive disabilities who have difficulties with speech. Tests conducted with six students in school settings identified the effectiveness of the use of technology to improve

196

P. Pachacama et al.

children’s daily communication, however it is still a challenge to develop solutions that support the end users according to their different needs. For people with cognitive disabilities, in many cases it is a challenge to get a job, as well as remembering routines, performing tasks. Taking this into consideration the work developed by Chang et al. [5] present a proposal for a task-pulse system to increase the independence of people with cognitive disabilities in workplaces by means of task prompting with the use of a PDA and beacons that facilitate their location, the tests of the system were carried out with eight people inside a community cafeteria operated mostly by people with cognitive disabilities where it is shown that the workload of the humandevice interface is low and the abilities that help in the performance of the tasks are high and reliable. In another study conducted by Cuascota et al. [3], emphasis is placed on the execution of daily tasks of people with cognitive disabilities and it is proposed the development of a mobile application called TK-2 where a gamma module is implemented to take the data obtained from the execution of sub-tasks of the predetermined activity in the application (origami). Jadan et al. [6] conducted a case study on a structured method in a web-based repository to assist in the literacy of children with special educational needs, the virtual repository called Tic@aula that allows for the creation and sharing of learning objects based on the IEEE LOM standard: i) literacy was carried out without the use of technology where the teacher labels common objects such as table, chair with the objective that the girl manages to identify and recognize the object, and ii) using the Tic@ula and a tablet, cards with the image and text of the object were shown. Finally, the authors conclude that the structured method in a web repository has a potential impact on the literacy process of children with special needs. Of the works mentioned, the solutions proposed in each one is focused on predefined activities; besides that, they do not focus their efforts on providing a tool that allows tutors to adapt the application for other types of activities based on the conditions of people with cognitive disabilities, which is why it is proposed as an added value the creation of an authoring tool that facilitates tutors the management of learning objects.

3 Implementation This section presents the architecture and development proposal of the author’s tool called TK-System. 3.1 Architecture The author’s tool proposed in this research work is based on the TK-2 application [3]. TK-System aims to provide tutors of people with cognitive disabilities a tool that provides the facility to: manage learning objects, manage the registration of students and display statistical graphs of the results obtained through TK-2 after the execution of the activities. For the development of the authoring tool node.js technology was used, for the design of the HTML and CSS interfaces, while to develop the functionalities of the application the Javascript programming language was used and as a MySql database manager where user information, multimedia elements and results obtained from each student are stored (see Fig. 1).

Authoring Tool for the Creation of Learning Objects

197

Fig. 1. TK-system architecture.

The architecture of the authoring tool consists of a server for the website from which you can manage the learning objects and users (tutors, learners), additionally a resource server is built for the TK-2 application, which allows to communicate the database with the mobile application, allowing to implement a user login and consume the resources of the learning objects created on the website. In the mobile application, the results of time, errors and help of the different users in each activity are obtained so that finally these are stored in the database through the resource server and in turn can be viewed on the website. All this architecture is hosted in an EC2 instance of AWS. 3.2 Interfaces A user-centered and easy-to-understand design has been used to create the interfaces, allowing the user to perform actions in a simple and intuitive manner. The main aspects that have been considered for the design are: Structure. The global structure of the site is generic and simple, it is designed based on the content and ordered in a sequential way. Navigation. The navigation bar located at the top contains links to the different components of the tool, this facilitates the intuitive movement of the tutors within the website. Colors and Styles. The selection of colors was made based on the concepts established in color psychology [7], to create a professional and pleasant environment for users who will make use of the tool (see Fig. 4).

3.3 Components and Functionality The architecture of Tk-System allows the tutors (see Fig. 2) create learning objects by activity through the web page, at the same time these activities are reflected in the mobile

198

P. Pachacama et al.

application to which the students will access by means of a user and password, once the execution of some activity is finished automatically the results (time, errors, helps) obtained are saved in the data base and are visible in the option of results in the web page through statistical graphs and summary tables.

Fig. 2. Overview of Tk-System functionality

The authoring tool is composed of several components, each containing subcomponents according to the required functionality (see Fig. 3). General Components. It consists of a Login from which the tutors must authenticate themselves with their credentials to enter the tool, then the main screen is displayed with a welcome message and a navigation bar with the following options to access other components of the website: home, users, activities, my account and logout. The option “My account” presents a submenu with the options i) My data that contains the information of the tutor’s name, surname, occupation and user, ii) Change password allows tutors to modify their password to access the tool by entering the current password and a new one. User Components. This component allows tutors to manage their users (students) within the tool, to create a new student is requested information of name, surname, date of birth, clinical diagnosis and a user to access the mobile application, then will be displayed in a table the list of students entered, each with the options to modify and delete. Activity Components. This component allows tutors to manage the learning objects (activities), it is composed of the following elements: 1) Add activity: it shows a form where it requests information of the name, image and a brief description of the activity, 2) Table with the list of the created activities, each one with its respective image, name, description and 3) Options to edit activity, steps of the activity, view results of the activity and delete activity (see Fig. 4) (Table 1).

Authoring Tool for the Creation of Learning Objects

199

Fig. 3. Components of the tool

Fig. 4. Activity component interface

Table 1. Accuracy levels of results by activity. Level

Description

General Shows the overall average of all users By user Displays the average of all attempts per user Specific Shows the results of a specific attempt

4 Evaluation The tests are carried out with the aim of identifying whether TK- System is an easy-touse tool for the creation of learning objects and to obtain recommendations to improve it. 4.1 Defining Metrics To measure the usability of the tool, the criteria of efficiency, effectiveness and user satisfaction, proposed in the usability framework of ISO 9241-11 [8] are considered

200

P. Pachacama et al.

each criterion has a percentage of 33.3% and the sum of the three represents the total usability, see Table 2. Table 2. Evaluation criteria. Criteria

Efficiency

Efficacy

Satisfaction

Measure

Time to perform a task

Number of errors made System Usability Scale when performing a task (SUS) [9]

Percentage of usability

33,33%

33,33%

Acceptance percentages

>11,1% Deficient 11,2–2,2% Acceptable = 38

ON

LOW

HIGH

Don’t Open Door

2.2 Flow Chart of Thermometer Operation Figure 2 shows the configuration process for the sensors network programming in the Arduino IDE software. 2.3 Infrared Sensors MLX90614 The MLX90614 is a sensor based on infrared light, making it possible to obtain the temperature of an object, with the advantage that it is not necessary to have direct contact with it. It has high accuracy of 0.5 °C over a wide temperature range (0.. + 50 °C) for both ambient temperature and object temperature. The MLX90614 is a sensor working with an inside state machine, which is devoted to controlling the measurements and calculating the temperature, then being visualized through the output PWM or SMBus [8].

258

S. Costanzo and A. M. Flores

Fig. 2. Flow chart of thermometer operation.

2.4 Influence of Environment Factors on the Operation of Infrared Sensors When measuring the human body temperature by means of infrared sensors in a given environment, not only the body’s own radiation is considered, but also other forms of radiation energy, such as the energy of the ambient temperature. In Fig. 3, it is possible to observe the reflected energy (R) caused by the ambient temperature, which has some effect on the precision of the body temperature measurements [9]. Environment Factors Incident Energy (I) Reflected Energy (R) Radiation Energy (E)

Fig. 3. Reflected energy caused by the ambient temperature on human body.

IoT Non-contact Body Temperature Measurement System

259

3 System Design and Implementation The access control system developed in this work is based on the measurement of body and ambient temperatures; its implementation starts from the design of a sensor network on Arduino board, which is made up of the MLX90614 infrared sensor used for noncontact body temperature measurement, and the DHT11 sensor used to measure the ambient temperature and the relative humidity. As an actuator of this sensors network, a servomotor is adopted, which is responsible for controlling access to a certain place based on the instructions sent by the Arduino to be able to open or close a door, that is, to allow or deny people access, based on the measured temperature ranges. The diagram of the sensors network can be seen in Fig. 4.

Fig. 4. Sensors network diagram.

4 Experimental Results 4.1 Communication with BLE Protocol BLE is a wireless personal network technology, whose main advantage is in the very low power requirements, acceptable range of communication, and interoperability in the world of chipset manufacturers [10]. In Fig. 5, the data sent to the raspberry pi on the Arduino serial monitor is illustrated.

260

S. Costanzo and A. M. Flores

Fig. 5. Data sent to the raspberry pi.

To receive data from Arduino on the Raspberry pi, it is necessary to install the “Bluetooth GATT SDK” library for Python, which allows to implement and communicate with any BLE device having a GATT profile [11]. After establishing the connection between the BLE of the Arduino and the BLE of the Raspberry Pi, data captured by the sensors are received and stored in a text file, whose format is shown in Fig. 6.

Fig. 6. Data format for the storage of information received in the raspberry pi.

4.2 Communication with the HTTP Protocol In order to transfer the information from the raspberry pi to the web platform, the HTTP communication protocol is adopted, by installing the “httplib2” library in Python that provides forms and methods to access web resources through HTTP [12]. This library is used to be able to separate each one of the sensors parameters found in the.txt document, using the httplib2 library, as shown in Fig. 7, and to send data to the web platform through the library “Requests”. To achieve a complete evaluation of the proposed IoT platform, a comparison is made between data obtained from the implemented system and those captured with a traditional body temperature thermometer. Experimental data refers to the temperature

IoT Non-contact Body Temperature Measurement System

261

Fig. 7. Sending data to the ThingSpeak web platform.

values measured on the same person, but at different times in a day. A distance equal about to 10 cm is considered between the detection system and the human body, specifically the forehead of the person. A calibration step procedure, at different ambient conditions, is applied prior to the measurement process. The achieved results, before and after the calibration step, are reported in Table 2 and Table 3, respectively. Table 2. Measurements with the implemented system and a traditional thermometer at different ambient temperatures and relative humidity (no calibration). Implemented system

Traditional thermometer

Temperature differences ° C

Ambient temp °C

Relative humidity %

Body temp °C

Body temp °C

10

98

18.01

35

16.99

15

88

22

35.36

13.36

22

60

25.47

35.4

9.93

24

75

28.15

35.72

7.57

29

67

31.12

36.5

5.38

34

44

33.3

36.6

3.3

36

47

33.72

36.76

3.04

Table 3. Measurements of the implemented system and traditional thermometer after performing the system calibration. Hour

Implemented system

Traditional thermometer

Absolute error

Relative error

Body Temp °C

x = x − x0

ε=

Body temp. °C Ambient temp. °C

Humidity %

6:00

35.55

22

61

35.4

0.15

0.42%

10:00

36.24

28

43

36.4

0.16

0.44%

14:00

36.99

34

31

37.2

0.21

0.56%

16:00

37.09

35

29

37

0.09

0.24%

18:00

36.41

29

60

36.6

0.19

0.52%

20:00

36.27

27

55

36.4

0.13

0.36%

0.075

0.42%

Average error

|x| x0 ∗ 100

262

S. Costanzo and A. M. Flores

The infrared sensor can present errors in the measurements due to some factors such as: – interference from outside light to the system; – ambient temperature, the lower the ambient temperature, the greater the impact on the system; – measurement distance: the field of view of the MLX90614 sensor must be considered, which in this case is equal to 35°, giving a measurement distance equal about to 10 cm between the person’s forehead and the thermometer.

Fig. 8. Temporal variation of measurements made on the ThingSpeak Platform: body temperature (a), ambient temperature (b) and relative humidity (c).

IoT Non-contact Body Temperature Measurement System

263

4.3 Information Display on the Web Platform Figure 8 shows the information appearing on the web platform, in terms of temporal variations of body temperature, ambient temperature and relative humidity. The ThingSpeak platform for data monitoring can also be accessed from the mobile phone through the ThingShow application, available in the Play Store; the connection with this application is made through the ID: 1112646 of the ThingSpeak Public Channel. Figure 9 shows the data sent from the sensor network to the web platform.

Fig. 9. Information display in the ThingSpeak Android application.

5 Conclusion An IoT thermographic system including infrared body temperature, ambient temperature/ humidity and servomotor sensors, has been presented in this work. The proposed IoT platform can be successfully adopted for access control of people into popular places through temperature tracking, thus preventing the spreading of pandemic situations, such as COVID-19. The inclusion of ambient temperature and relative humidity detection provides additional information to improve the accuracy of infrared body thermographic measurement, due to the influence of environmental factors. Data acquisition and access control are managed through a web platform, leading to a compact, versatile and low-cost system. Experimental validations have been discussed, by obtaining an average relative error equal about to 0.42%, as compared to a traditional thermometer, thus demonstrating the reliability of the proposed IoT system. The actual prototyped system presents some limitations, due to the non-linearity the body temperature scale, and the low temporal rate updating of Thingspeak platform. The use of the medical version of the MLX90614 infrared sensor is proposed as future work, which according to the data sheet has an accuracy of ± 0.02 °C. Another future

264

S. Costanzo and A. M. Flores

advancement will be the replacement of Arduino board with the ESP32 board widely used in IoT applications, which incorporates a BLE module and a Wi-Fi module, thus allowing data to be directly sent to the web platform with ultra-low power consumption, by giving at the same time system miniaturization and cost reduction.

References 1. Thermal cameras are being outfitted to detect fever and conduct contact tracing for COVID19. IEEE Spectrum. https://spectrum.ieee.org/news-from-around-ieee/the-institute/ieeemember-news/thermal-cameras-are-being-outfitted-to-detect-fever-and-conduct-contact-tra cing-for-covid19 2. Zhang, J.: Development of a non-contact infrared thermometer. In: International Conference Advanced Engineering and Technology Research (AETR 2017), vol 153, p. 311 (2017) 3. Vulpe, A., Lupu, C., Mihai, C.: Research on infrared body temperature measurement – virus spreading prevention. In: 2020 12th International Conference on Electronics, Computers and Artificial Intelligence (ECAI), Bucharest, Romania, vol 1, pp. 1–4 (2020) 4. Wijaya, N., Oktavihandani, Z., Kunal, K., Helmy, E., Nguyen, P.: The design of tympani thermometer using passive infrared sensor. J. Robot. Control (JRC) 1, 1–4 (2020) 5. Giddings, S.B.: Hawking radiation, the Stefan-Boltzmann law, and unitarization. Phys. Lett. B 754, 40–42 (2016) 6. Lahiri, B.B., Bagavathiappan, S., Jayakumar, T., Philip, J.: Medical applications of infrared thermography: a review. Infrared Phys. Technol. 1, 226 (2012) 7. Usamentiaga, R., Venegas, P., Guerediaga, J., Vega, L., Molleda, L., Bulnes, F.G.: Infrared thermography for temperature measurement and non-destructive testing. Open Access Sens. 1, 12308–12310 (2014) 8. MELEXIS Microelectronic Integrated Systems, Electronic Components Datasheet Search. https://www.alldatasheet.com/datasheet-pdf/pdf/551163/MELEXIS/MLX90614.html 9. Wei, S., Qin, W., Han, L., Cheng, F.: The research on compensation algorithm of infrared temperature measurement based on intelligent sensors. Springer 1, 2–3 (2018) 10. BLE (Bluetooth Low Energy), Innovation in Lighting Technology. https://www.elt.es/ble-blu etooth-low-energy. 11. Bluetooth GATT SDK for Python, GitHub, Inc. https://github.com/getsenic/gatt-python#ins talling-gatt-sdk-for-python. 12. Python httplib2 – HTTP GET and POST examples, HowToDoInJava. https://howtodoinjava. com/python/python-library/httplib2-http-get-post-requests/

Miscellaneous from Workshops

Gamification in MOOC Context. Documentary Analysis Angel Torres-Toukoumidis1(B) , Sonia Esther González-Moreno2 , and Jesús Manuel Palma-Ruiz2 1 Universidad Politécnica Salesiana, Cuenca Azuay 010105, Ecuador

[email protected] 2 Universidad Autónoma de Chihuahua, 31300 Chihuahua, Mexico

Abstract. Massive Online Open Courses (MOOC) have been set up as an alternative for teaching during the pandemic. Therefore, the current situation of this emerging technique is analyzed through documentary analysis, deepening in the general gamification characteristics of the courses offered in 10 MOOC platforms. The findings obtained have allowed to organize them into 4 groups of courses on gamification, showing that although conceptually it is centered in the definition of Deterding et al. (2011), there is still some confusion with respect to other definitions such as serious games and videogames; Besides, it is also prescribed that beyond courses entirely directed towards gamification, several disciplines, among them, computer science, education and business administration have used in their content the realization of particular units towards gamification. In conclusion, MOOCs have clearly served to increase the visibility, presence, and teaching of emerging techniques, promoting interest and understanding of the subject by users immersed in the digital ecosystem. Keywords: Gamification · MOOC · Learning · Emerging technologies · Digitalization · Platforms

1 Introduction Due to the health crisis produced by the pandemic, non-formal education represented through online courses has been envisioned as an educational alternative, which has reflected an exponential increase in new student registrations on these platforms [1]. The Massive Online Open Courses (MOOC) have been studied in depth within the academic community, reviewing design typologies [2], user profiles [3], combined with different learning methodologies [4], accessibility [5] and perception [6].However, on this occasion, it is considered as a premise its involvement in the development of gamification, a trend that has permeated different areas of knowledge such as education [7], administrative sciences [8], marketing [9], etc. Although there is evidence that gamification has been implemented in more MOOC platforms to improve the commitment [10] and participation of students [11] generating positive results in the learning experience. The approach of this research concentrates © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 267–277, 2021. https://doi.org/10.1007/978-3-030-72654-6_26

268

A. Torres-Toukoumidis et al.

on analyzing a different edge, focusing on exploring the offer and the exposed content of the gamification courses proposed in the MOOCs. This problematization arises because nowadays gamification is exhibited as an adaptable resource to be taught from different contexts, where we can observe faculties of computer science, design, marketing, education offering general courses about the use of game elements in non-playful environments or, failing that, proposing some unit related to specialized courses where gamification is involved. From there several questions arise: What kind of content is taught about gamification in these courses? What concepts and ideas of gamification are the most widespread? Is there a theoretical uniqueness in the gamification courses? How important are the practical cases for teaching gamification? What role plays the digitalization of gamification experience in the contents taught? The gamification, as it happens with all those emerging technologies, becomes a favorable space for conceptual heterogeneity, diversifying in such a way that the vertebral construction of ideas is blurred, having repercussions a posteriori in the formative content presented in the courses. For example, the concept of gamification contains several general definitions [12] and [13] that differ from each other. This is even more so when the concept is determined by its etymology [14], or when notions specific to education [15], business administration [16] and information technology [17] are involved. These disjunctives demonstrate that, in effect, the innovative character of this research is oriented toward a detailed review of the composition of the training courses that are carried out on gamification or that this new strategy indirectly encompasses in their content.

2 Context of Gamification in MOOC A typical MOOC is executed through an agreement in which the educational sector outsources the external services of providers of multimedia materials [18]. Reisman (2014) [19] assessed that these types of courses are based on the well-known Learning Management Systems (LMS) such as Blackboard or Moodle, adjusted to increase the number of students. LMS are considered pedagogically effective and supportive of satisfactory teaching practices [20]. In other words, MOOC are online courses offered on an LMS platform that has been expanded, and to which some social media tools are also added. On the earlier years of MOOC – the most important and used platforms were Udacity, Coursera and edX [21]. According to data presented by Class Central, an online directory that lists MOOC courses and is a comprehensive source of information on the subject [22] in 2015 more people enrolled in MOOC than in the first three years of the movement [23], Dhawal Shah, founder and CEO of Class Central, estimated that based on the number of courses offered in 2015 the main platforms were Coursera, edX and Canvas Network. However, progressively more universities and institutions have joined this trend and have incorporated various types of software, design, and functionalities [24]. At the beginning of 2018, Shah (2017) [25] indicated that based on the number of students enrolled in their courses in 2017, the most relevant platforms were Coursera, edX and XuetangX; the latter is a Chinese platform developed by Tsinghua University. Table 1 shows the five highest-ranking platforms (based on the number of registered students and courses offered) for the years 2017, 2018, and 2019.

Gamification in MOOC Context. Documentary Analysis

269

Table 1. Highest-ranking platforms offering MOOC, according to class central. Classification

2017

2018

2019

1st place

Coursera

Coursera

Coursera

Number of students 30 million

37 million

45 million

2nd place

edX

edX

18 million

24 million

edX

Number of students 14 million 3rd place

XuetangX

XuetangX

Udacity

Number of students 9.3 million

14 million

1.5 million

4th place

Udacity

FutureLearn

10 million

10 million

Udacity

Number of students 8 million 5th place

FutureLearn FutureLearn Swayam

Number of students 7.1 million

8.7 million

10 million

Knowing this, the problem that concerns this research, seeks to recognize the influence of gamification on MOOCs. The general objective of this study is examining the linkage between gamification courses and MOOC platforms. To this purpose, the following specific objectives were established: analyzing the overall characteristics of gamification MOOC courses; and determining the content prescribed by gamification modules.

3 Methodology This study followed a qualitative research methodology. Considering a systematic approach for reviewing computer-based and Internet-transmitted material, the data was acquired through both documental analysis and non-participant observation; later the data was evaluated using a content analysis approach. Figure 1 depicts an overview of the process followed in this research. 3.1 Documentary Analysis Documentary analysis is a form of qualitative research that methodically examines published material; either in a printed or electronic presentation [26]. On this subject, indicates that documents are a valuable tool that help understand the research phenomena, establishing the accounts and background of an environment, the situations generated, and the regular and abnormal functioning of the studied phenomena. The document analysis approach includes the following phases: document sourcing, document selection, data collection and management, and data analysis. The first step, document sourcing focused on searching structured data through Internet engines in making information accessible [27]. For this study, the researchers considered two different but complementary sources to retrieve the MOOC in gamification. The first option was to access each of the eight platforms offering MOOC. Within each of these platforms, the authors

270

A. Torres-Toukoumidis et al.

Results

2. Non-participant observation

1. Document Analysis

1.1 Document sourcing

1.2 Document selection

Source 1: Class Central

Keywords: gamification, "gamificación", and "ludificación"

Source 2: MOOC platforms

Inclusive criteria: courses in Spanish or English

1.3 Data collection and management

Categories: (institution, country, duration, languages, level, objective, rating, and URL)

1.4 Data analysis

Classification: 1) Available MOOC about gamification, 2) Closed MOOC about gamificacion, 3) Available MOOC with a partial study of gamification, and 4) Closed MOOC with a partial study of gamification

Excluded criteria: payment required or subject not relevant to the research

Fig. 1. Overview of the research methods

performed keyword searches for MOOC containing the keyword “gamification” (and its Spanish translation “gamificación” or “ludificación”). In most cases, the platforms retrieved a list of results that either contained those keywords in the title of the MOOC, as a descriptor of the course, or as a subtitle of one or more of the modules or sections within the course. The second option considered the Class Central directory (classcentral.com) to perform the same keyword searches containing the word “Gamification” (and its translations into Spanish: “Gamificación” or “Ludificación”). The results displayed general information about the MOOC, including the title of the course, the institutions and the platform hosting the course and its corresponding URL. This search technique took place from September 20th to the 30th , 2020. The second phase, document selection, the authors created a set of inclusionary and exclusionary criteria to identify and retrieve those courses that would better fulfill the scope of the study [28]. The inclusionary criteria were courses containing the keyword “Gamification” (and its translations in Spanish “Gamificación” or “Ludificación”) that were either available or closed at that time. Excluded criteria involved courses that

Gamification in MOOC Context. Documentary Analysis

271

Table 2. Number of MOOC that stemmed from the steps: document sourcing and document selection. Coursera EdX Udacity FutureLearn Swayam XuetangX MiríadaX Iversity Class Central Number of 11 MOOC with any keywords

10

51

0

0

0

4

2

25

2

3

0

0

0

0

2

2

10

Percentage 82 of rejection

70

100

–

–

–

50

0

64

Number of MOOC after “document selection”

– although identified as MOOC – required a payment for enrollment. Additional inclusionary criteria were integrated depending on the characteristics of the platforms. For example, Coursera subdivides the courses into three categories (open courses, guided projects, and specializations); only “open courses” were included in the search because guided projects and specializations include fees. One interesting finding was that 73% of the MOOC retrieved during the Document Sourcing search were rejected after a thorough inspection because they were either not related to the study of gamification or required payment for enrollment (Table 2). A second interesting observation was that FutureLearn did not yield any MOOC with the keyword “gamification” but listed 22 MOOC related to “game”; from those results, nine were MOOC associated specifically with videogames (development, history, design, starting a business and data science). The third phase, data collection and management, means interpretation of any data requires reasoning and acknowledgement of the original intended meaning. In this study, the authors established that the purpose of the information shown through the platforms was to convince potential students to enroll in the MOOC; therefore, the information was concise and designed from a marketing standpoint. Thus, it presented, from origin, a pre-pattern of categories: name of the course, type of MOOC, offered by (institution), country, names of the instructors, duration, languages, level, objective, rating, and URL. During this phase, the authors also agreed on a classification [29] that later assisted them to interpret the information. The four classifications were: a) b) c) d)

Group 1: Available MOOC about Gamification. Group 2: Closed MOOC about Gamification. Group 3: Available MOOC with a partial study of Gamification. Group 4: Closed MOOC with a partial study of Gamification.

While identifying the MOOC, the authors cross-referenced those found within the eight platforms with the Class Central directory to corroborate the results. Nine MOOC were identified directly from the platforms, which were also listed in Class Central.

272

A. Torres-Toukoumidis et al.

However, an additional MOOC called “Instructional design and technology: learning theories” from edX was identified in Class Central and added to the final results. Accordingly, MOOC were organized into one main listing to proceed with the data analysis as described in Table 3. Table 3. Group classification of MOOC after data collection and management. Group

Name of the MOOC

Group 1: All available MOOC about gamification

Gamification (Coursera)

Group 2: Not available MOOC about gamification

Accessible gamification (edX)

Leading Change: Go beyond gamification with gameful learning (edX) Introducción a la gamificación a través de casos prácticos (Introduction to gamification through practical cases) - 5th edition (MiríadaX) Gamification design 2015 (Iversity)

Group 3: Available MOOC that partially teach Claves para la innovación en la docencia about gamification universitaria (Keys for innovation in university teaching) (Coursera) Tecnologías web emergentes para la enseñanza virtual (edX) Instructional Design and Technology: Learning theories (edX) Corporate digital learning (Iversity) Group 4: Not available MOOC that partially teach about gamification

Diseño, organización y evaluación de videojuegos y gamificación (Design, organization and evaluation of videogames and gamification) (MiríadaX)

The last and fourth phase, data analysis, it was conducted in a “way that empirical knowledge is produced and understanding is developed” (Bowen, 2009, p. 34). To analyze the content of the identified MOOC that fulfilled the inclusionary criteria, the information was organized into the pre-pattern of categories and the four-group classifications described in the data collection and management. The authors implemented a non-participant observation technique to reach a more complete and sturdy analysis of MOOC platforms. It is worth emphasizing that the documentation analysis was conducted from March 25th to October 21st , 2020.

4 Results The first part of this section describes the results for each of the group classifications. The second part offers the non-participant observation highlights conducted on the two

Gamification in MOOC Context. Documentary Analysis

273

Table 4. List of available MOOC about Gamification (Group 1) Platform

Name

Type and duration

Provider

Language

Coursera

Gamification

Course 17 h

University of Pennsylvania (United States)

English (subtitles in French, Ukrainian, Simplified Chinese, Brazilian Portuguese, Vietnamese, Russian, Turkish, English, Spanish and Kazakh)

edX

Leading Change: Go beyond gamification with gameful learning

Course 20 h

University of Michigan (United States)

English (transcripts in English)

MOOC that give a full emphasis to gamification (Group 1), and the four MOOC that partially taught gamification on a specific module (Group 3). Group 1 contained all available MOOC about gamification. In these courses the registration process was open, and enrolled students would be able to visualize their progress. Additionally, in the case of Coursera, the students enrolled would receive a grade and feedback of their activities and projects. Table 4 contains the two MOOC included in Group 1; one offered by Coursera and another by edX. Both MOOC are offered by universities from the United States with English as the official language; although Coursera offers subtitles in ten languages, whereas edX only has transcripts in English. Table 5. List of closed (not available) MOOC about Gamification (Group 2) Platform

Name

Type and duration

Provider

Language

edX

Accessible gamification

Course 12.5 h

Georgia Tech (United States)

English (transcripts in English)

MiríadaX

Introducción a la gamificación a través de casos prácticos (Introduction to gamification through practical cases) - 5th edition

Course 20 h

Universitat Oberta Spanish de Catalunya (Spain)

Iversity

Gamification design 2015 Course 10 weeks

Springer Nature (international)

English

274

A. Torres-Toukoumidis et al.

Group 2 enclosed the not available MOOC about gamification. Table 5 shows the three MOOC included in this classification. The first one was supported in the edX platform and offered in English by Georgia Tech from the United States. Even though the registration process is closed, interested students are able to examine materials without receiving any feedback or open communication from the instructors. The MOOC offered by MiríadaX had specific sign-up dates; when closed, potential students are not able to review the content from the different modules. The only information available is a video presentation on the website; where it can also be observed that the course is taught in Spanish by the Universitat Oberta de Catalunya (Open University of Catalunya), and required 20 h for completion. The last course in this group classification was offered by Iversity. That MOOC was offered in 2015 and is no longer open for registration, nor available to examine materials or browse the educational videos. Table 6. List of available of MOOC that partially teach about Gamification (Group 3) Platform

Name

Coursera

Type and duration

Provider

Language

Unit

Claves para la Course innovación en 12 h la docencia universitaria

Universitat de Barcelona (Spain)

Spanish

Week 7 is a module called “Gamification”

edX

Tecnologías web emergentes para la enseñanza virtual

Course 18 h

Universidad Galileo (Guatemala)

Spanish

Lesson 2: contains information about MOOC and gamification

edX

Instructional Design and Technology: Learning theories

Course 40 h

University of Maryland (United States)

English

Week 6 is called “Learning approach models: experiential learning, project-based learning, and gamification”

Iversity

Corporate digital learning

Course 4 modules

KPMG (international)

English

Chapter 3 (unit 3) contains a module called “Gamification in corporate digital learning”

Group 3 contains the available MOOC that partially address a subject or a module about gamification. Table 6 contains the four MOOC included in this group classification; one is available through Coursera in Spanish, two by edX (one in Spanish and one in

Gamification in MOOC Context. Documentary Analysis

275

English), and one by Iversity in English. The MOOC offered by Iversity is managed by KPMG corporation, an international network of accounting, taxing, and auditing firms with worldwide presence. The MOOC from this Group 3 could be further divided into two areas: MOOC within the educational industry, and MOOC with a business objective. As such, the approaches given to gamification widens. In addition, the same table contains the specific location (or week) of the gamification module in each of the four courses. Finally, Group 4 contains the MOOC that partially address a subject or a module about Gamification but are currently closed. Only the MOOC “Diseño, organización y evaluación de videojuegos y gamificación” (Design, organization and evaluation of videogames and gamification) was included in this group. Although this particular MiríadaX MOOC indicated that students could browse through the material without any expectation of interaction, the authors were unable to sign-up or access the contents of this course. However, its description indicates that the MOOC is a six-week Spanish-speaking content offered by Universidad Europea from Spain with an emphasis in computer science.

5 Conclusions The following research has summarized multiple significant advances, in which, in addition to answering the research questions, the specific objectives determined during the research are also met. Firstly, it is evident that gamification has been extended through the main MOOC platforms, and it is possible to infer that gamification is oriented to formalize the basic and general knowledge of this technique. Nevertheless, courses have also been built to teach disciplines such as education, computer science and business administration, which incorporate in one of their units the teaching of gamification in its theoretical development. In short, gamification is one of the emerging trends that, catalyzed by the influence of new information and communication technologies have managed to establish themselves in digital training spaces, forming new ways to involve and attract more users in their development. Similarly, it is worth mentioning that although the concept of Deterding et al. (2011) has been established as a conceptual pillar of gamification, some courses have been confused in combination with other concepts such as serious games and videogames. While the teaching of gamification through the MOOC is denoted as a didactic path, the situation produced by the pandemic could involve its digital prescription in greater detail, especially by including elements of play during the learning sessions. In fact, it would be worthwhile to see if other didactics such as inverted classroom, project based learning, problem based learning contain such frequency in the MOOC platforms. In short, the authors propose 2 recommendations, the first one that gamification courses presented in MOOC platforms should present clearly and transparently the information of the subject and the approach, in that way, the users will be able to select with more diligence the course that more adapts to their needs and in second instance, It is recommended that researchers proceed with future cyber-ethnographic analyses that will allow them to know more closely the didactics applied in the teaching of the MOOCs on gamification, as well as to know the repercussion and impact of these MOOCs on gamification on the users, evaluating their feedback and use in their environment.

276

A. Torres-Toukoumidis et al.

References 1. Quijano-Escate, R., Rebatta-Acuña, A., Garayar-Peceros, H., Gutierrez-Flores, K.E., Bendezu-Quispe, G.: Aprendizaje en tiempos de aislamiento social: cursos masivos abiertos en línea sobre la COVID-19. Rev. Peru. Med. Exp. Salud Publica 37(2), 18–45 (2020) 2. Milligan, S.K., Griffin, P.: Understanding learning and learning design in MOOCs: a measurement-based interpretation. J. Learn. Anal. 3(2), 88–115 (2016) 3. Watson, S.L., Watson, W.R., Yu, J.H., Alamri, H., Mueller, C.: Learner profiles of attitudinal learning in a MOOC: an explanatory sequential mixed methods study. Comput. Educ. 114, 274–285 (2017) 4. Klemke, R., Eradze, M., Antonaci, A.: The flipped MOOC: using gamification and learning analytics in MOOC design—a conceptual approach. Educ. Sci. 8(1), 25–38 (2018) 5. Larionova, V., Brown, K., Bystrova, T., Sinitsyn, E.: Russian perspectives of online learning technologies in higher education: an empirical study of a MOOC. Res. Comp. Int. Educ. 13(1), 70–91 (2018) 6. You, H.: Students’ perception about learning using MOOC. Int. J. Emerg. Technol. Learn. (iJET) 14(18), 203–208 (2019) 7. Torres-Toukoumidis, A., Romero, L.: Aprender jugando. La gamificación en el aula. In: García, R., Perez, A., Torres-Toukoumidis, A. (eds.) Educar para los nuevos Medios, Claves para el desarrollo de la competencia mediática en el entorno digital, pp. 61–72. Editorial Universitaria Abya-Yala, Quito (2018) 8. Torres-Toukoumidis, A., Marín-Mateos, P.: Gamificación en aplicaciones móviles para servicios bancarios de España. Retos. Revista de Ciencias de la Administración y Economía 7(13), 43–57 (2017) 9. Huotari, K., Hamari, J.: A definition for gamification: anchoring gamification in the service marketing literature. Electron. Mark. 27(1), 21–31 (2017) 10. Morales, M., Amado-Salvatierra, H.R., Hernández, R., Pirker, J., Gütl, C.: A practical experience on the use of gamification in MOOC courses as a strategy to increase motivation. In: International Workshop on Learning Technology for Education Challenges, pp. 139–149. Springer, Cham (2016) 11. Chang, J.W., Wei, H.Y.: Exploring engaging gamification mechanics in massive online open courses. J. Educ. Technol. Soc. 19(2), 177–203 (2016) 12. 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. ACM (2011) 13. Yohannis, A.R., Prabowo, Y.D., Waworuntu, A.: Defining gamification: from lexical meaning and process viewpoint towards a gameful reality. In: Information Technology Systems and Innovation, pp. 284–289. IEEE (2014) 14. Torres-Toukoumidis, A.: Evaluación de políticas públicas con técnicas de gamificación para la educación ciudadana. Universidad de Huelva, Spain (2016) 15. Kapp, K.M.: The Gamification of Learning and Instruction: Game-Based Methods and Strategies for Training and Education. Wiley, New York (2012) 16. Zichermann, G., Cunningham, C.: Gamification by Design: Implementing Game Mechanics in Web and Mobile Apps. O’Reilly Media Inc, Massachusetts (2011) 17. Huotari, K., Hamari, J.: Defining gamification: a service marketing perspective. In: Proceedings of the 16th International Academic MindTrek Conference, pp. 17–22. ACM (2012) 18. Baggaley, J.: MOOC rampant. Distance Educ. 34, 368–378 (2013) 19. Reisman, S.: The future of online instruction, part 2. Computer 47(6), 82–84 (2014)

Gamification in MOOC Context. Documentary Analysis

277

20. Palma-Ruiz, J.M., González-Moreno, S.E., Cortés-Montalvo, J.A.: Sistemas de gestión del aprendizaje en dispositivos móviles: evidencia de aceptación en una universidad pública de México. Innovación Educativa 19(79), 35–56 (2019) 21. Matías González, H., Pérez Avila, A.: Los Cursos en Línea Masivos y Abiertos (MOOC) como alternativa para la educación a distancia. GECONTEC: revista Internacional de Gestión del Conocimiento y la Tecnología 2, 41–49 (2014) 22. Cisel, M.: The structure of the MOOC ecosystem as revealed by course aggregators. Am. J. Distance Educ. 33(3), 212–227 (2019) 23. Class Central. The Report. https://www.classcentral.com/report/moocs-2015-stats/. Accessed 21 Sept 2020 24. Poy, R., Gonzales-Aguilar, A.: Factores de éxito de los MOOC: algunas consideraciones críticas. Iberian J. Inf. Syst. Technol. (e1), 105–118 (2014) 25. Class Central. The Report. https://www.classcentral.com/report/moocs-2017-stats/. Accessed 26 Sept 2020 26. Bowen, G.A.: Document analysis as a qualitative research method. Qual. Res. J. 9(2), 27–40 (2009) 27. Coffman, J., Weaver, A.C.: A framework for evaluating database keyword search strategies. In: Proceedings of the 19th ACM International Conference on Information and Knowledge Management - CIKM 2010, pp. 729–753 (2010) 28. Frey, B.B.: The Sage Encyclopedia of Educational Research, Measurement, and Evaluation. Sage Publications, Thousand Oaks (2018) 29. Fereday, J., Muir-Cochrane, E.: Demonstrating rigor using thematic analysis: a hybrid approach of inductive and deductive coding and theme development. Int. J. Qual. Methods 5(1), 103–128 (2006)

RETRACTED CHAPTER: Semantic Analysis on Twitter of Two Religious’ Leaders: @Pontifex and @DalaiLama Angel Torres-Toukoumidis1(B) , Isidro Marín-Gutiérrez2 , and Javier Herrán Gómez1 1 Universidad Politécnica Salesiana, Cuenca, Azuay 010105, Ecuador

[email protected]

ER

2 Universidad Técnica Particular de Loja, San Cayetano Alto, Calle París, Loja, Ecuador

TE

D

C

H

A

PT

Abstract. Social media stands out as the main setting of interaction within the digital ecosystem permeating the communication of religious leaders. Admitting such a premise, the following research establishes a descriptive analysis of the interaction carried out by Pope Francis and Dalai Lama in their respective Twitter accounts determining variables such as frequency of publication, use of hashtags and culminating in the revision of the lexicon used in tweets through text mining software tools. The main results denote greater activity by the Pope on Twitter in terms of the frequency and quantity of messages; on the other hand, as far as the use of hashtags is concerned, the Pope’s account highlights events related to the religious doctrine he leads, while the Dalai Lama focuses on its application to current situations in the global situation. In the third objective, the lexicon incorporated by both in their messages differs. While the Pope focuses on promoting institutional identity, the Dalai Lama builds his tweets on general human values. In short, it can be concluded that the following research manages to deepen the paradigmatic axis by combining it with the dissemination in digital networks.

A

1 Introduction

C

Keywords: Twitter · Lexicon · Hashtags · Social media · Semantic · Leaders

R

ET

R

On October 8th, 2011, the Dalai Lama participated in Google+ with Archbishop Desmond Tutu to celebrate his 80th birthday [1]. It happened because the Dalai Lama was denied a visa to visit South Africa [2]. The Dalai Lama participated in the live hangout open to the public. Later a user posted images of the hangout on YouTube. The public praised the two Nobel Peace Prize winners for their decision to meet virtually despite government restrictions [3]. On January 23, 2014, Pope Francis described the Internet and its immense possibilities for meeting and solidarity as a gift from God [4]. It is necessary to study religion in social networks due to the growing dominance of religion in the media. Religion is a central component of human culture, identity, and social relations [5]. Prior to 2010, no international religious leader had posted content on Twitter. This all changed after one person supplanted the Dalai Lama on Twitter. After this incident, the The original version of this chapter was retracted: The retraction note to this chapter is available at https://doi.org/10.1007/978-3-030-72654-6_47 © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021, corrected publication 2022 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 278–286, 2021. https://doi.org/10.1007/978-3-030-72654-6_27

RETRACTED CHAPTER: Semantic Analysis on Twitter of Two Religious’ Leaders

279

R

ET

R

A

C

TE

D

C

H

A

PT

ER

real Dalai Lama joined Twitter in February of that year and other international religious leaders did the same. Religious organizations aim to serve their followers. Communication is crucial to this. Many religions require the consumption of multimedia materials through books, music and so on [6]. The greatest leaps in education and communication were with the invention of the printing press that inspired a global increase in literacy. Religious organizations are pioneers in terms of communication [7]. The emergence of the Internet and social networks is now important for faith-based organizations [8] meaning that in many parts of the world, major religions represent the primary focus of communication at local, national and global levels. Surveying religious leaders, it was found the use of Internet as an educational resource and that most spiritual leaders go online to promote their congregational worship [9]. Regarding the relationship between religion and social networks, 20% of Americans posted content about their religion on social networks. Internet users seek religious information and communication through the Internet [10]. While there are social network accounts for temples, synagogues, churches and monasteries, they are few and far between. Religious leaders are accepted as personified representations of religious organizations. Catholicism and Buddhism were specifically chosen for several reasons, (a) they are actively practiced internationally; (b) they have living religious leaders who are active in social networks; (c) they have a rich history of communication both outside and within the Internet; and (d) they have a senior religious leader [10]. His Holiness the 14th Dalai Lama, a Buddhist monk and spiritual leader from Tibet, was recognized within two years as the reincarnated. The Dalai Lama assumed all political and spiritual power in Tibet at age 15 and is known for his non-violent approach and the lifelong struggle for Tibet’s liberation from Chinese rule. The Dalai Lama is now 80 years old and has traveled the world encouraging peace initiatives. The Dalai Lama’s Twitter username is @DalaiLama and his first tweet was on February 22, 2010 [10]. Pope Francis is the first American Pope and was elected Pope at age 76 on March 13, 2013. Pope Francis is known for his humble approach to the papacy, his modern style towards the papacy, and his compassion for the poor. Pope Francis was considered the “Pope of the people” and is the second Pope to have a Twitter account. Pope Francis’ Twitter username is @Pontifex, which is Latin for “bridge builder. Pope Francis has nine Twitter accounts. His first tweet occurred on March 17, 2013, five days after he was elected Supreme Pontiff [10]. Religious leaders are perceived as building trust and relationships with their followers by speaking to them “personally”, without having a conversation with them through the traditional model of symmetrical communication. It is through communication on Twitter and personal references on Twitter that the Pope and the Dalai Lama are perceived positively and that helps to maintain a positive relationship between the religious leader and his followers, thus inspiring and encouraging commitment to their religion [11]. The presence of religious leaders in social media can serve not only religious teaching messages but also interreligious and ecumenical dialogue. An episode from late May 2013 demonstrates this. The official headquarters of the Church of England, Lambeth Palace, reiterated Pope Francis’ idea of prayer in his account @lambethpalace.

280

A. Torres-Toukoumidis et al.

1.1 Quantitative Analysis of Content on Twitter

TE

D

C

H

A

PT

ER

Twitter is a tool for measuring public opinion [12]. All tweets are public unless the account holder specifies, explaining that Twitter is an accurate and reliable snapshot of global sentiment [13]. Twitter is a web-based social networking site that was publicly launched in 2006. This social network allows its users to post photos, text, audio, video and information across the web. Twitter allows its users to include website URLs, hashtags (#), other accounts (@), photos, and videos of the published content. Twitter differs from other social networks by its 140 characters limit per tweet. Twitter is also considered a website or microblogging tool [10]. Twitter is a free resource that encourages users to produce and consume information. Twitter is available anywhere in the world with Internet access or a cell phone data plan. Twitter is available in 33 languages with 288 million active users per month. Every time a user tweets, or publishes content, the tweet can be seen by the public or only by approved “followers”, people who have been granted access by the account holder. Users post an average of 500 million tweets per day. The growth of active users on Twitter is exponential. The number of users increased by 28% between 2013 and 2014. A large part of the steady growth is attributed to the ease of use of the social networking site, which allows users from all walks of life to learn the jargon easily, regardless of their social status, educational level or geographic location [14]. Twitter is an excellent tool for academic research. Researchers can easily measure opinion, relevance, and attribution through content analysis of public tweets [15].

2 Methodology

ET

R

A

C

This article presents a perspective focused on a descriptive view of the interaction from religious leader and its influence through Twitter accounts to know the main contents of interest at a communicational level in each religious doctrine. For this purpose, the social network Twitter has been selected as the main source of the content exposed by 2 media leaders such as the Dalai Lama and Pope Francis, thus configuring 3 specific objectives:

R

– Determining the frequency of publications of both spiritual leaders on Twitter. – Identifying the most repeated hashtags of their publications on Twitter – Applying a sentiment analysis referring to the lexicon presented in the tweets of both leaders. The design used in this research contains a deductive-quantitative approach with a descriptive scope ordered by the application of content analysis, defined by Bardin (1996) as the set of techniques for the analysis of communications willing to obtain quantitative indicators by systematic and objective procedures of description of the messages allowing the inference of knowledge related to the conditions of production/reception within the social context of these messages. Specifically, the nature of social networks provides a significant construction of units of analysis that facilitate the coverage of the

RETRACTED CHAPTER: Semantic Analysis on Twitter of Two Religious’ Leaders

281

C

TE

D

C

H

A

PT

ER

messages through computational processes and algorithms that allow the collection of patterns arranged by the fluctuations categorized in Twitter [16]. For this purpose, two religious authorities have been selected, Pope Francis and Dalai Lama excluding leaders of other religions because both have important similarities: both act as examples for Catholicism and Buddhism respectively, both act as promoters of human values, and of course both use social networks to communicate their message. In this case, we have selected the official English accounts of Pope Francis @Pontifex, created in February 2012 which contains more than 18 million followers and the account of the Dalai Lama @DalaiLama created in February 2009 which maintains today more than 19 million followers. It should also be noted that the Pope has 8 more accounts in different languages, Latin, German, Spanish, Portuguese, Polish, Italian, French and Arabic which in total would imply more than 30 million followers. However, recognizing that the Dalai Lama only has one account in English, this study was oriented only to tweets written in that language. Although the Twitter accounts of the Pope and Dalai Lama, specializes in the rhetoric of the community, while the following study seeks to deepen the quantitative method from computer tools obtaining complementary and equally significant data on the sequential use of social networks and lexicography in religious doctrines [17]. The tweets analyzed are dated between January 2019 and July 2020, basically 19 months of publications made by both religious leaders were considered. The reason that supports such selection is that initially it was going to be reviewed only for one year, January 2019 until January 2020, however, the escalation produced by the global pandemic has been a phenomenon that was also intended to be explored during the analysis of content. For the collection of tweets, first, performed the authentication of Twitter to access the application programming interface (API) through the Software R version 4.02, then established a script in that software, which allowed modularizing the extraction of tweets referred to the objectives that contained the following components:

R

ET

R

A

Stringr ROAuth RCurl ggplot2 tm RJSONIO wordcloud gridExtra plyr score_sentiment.R func_on_tweet.R graphs.R

The Script codes exposed respond in computer language translated to the information obtained about the units of analysis: frequency of publication, distribution of tweets, labels and lexicon, being systematized from a sample estimated in tweets (n = 1,803) that reflects the terminological configuration of the publications issued by both religious

282

A. Torres-Toukoumidis et al.

leaders. Likewise, it should be expressed that the function of this study is not to make a subjective comparison between Pope Francis and the Dalai Lama nor to demonstrate direct traits of their personalities, but to demonstrate, in short, the interaction of their doctrine through a lexicological review of their respective tweets.

3 Results

A

C

TE

D

C

H

A

PT

ER

As for the preliminary results, the first figure shows the number of tweets made by both religious leaders during 19 months, in the case of Pope Francis’ account, @Pontifex, published 753 more than the Dalai Lama, @DalaiLama, generating a percentage difference of 556% more, meaning that, Pope Francis generated 5 times more publications than Dalai Lama (Fig. 1).

ET

R

Fig. 1. Number of tweets from @Pontifex and @DalaiLama.

R

Likewise, concerning the distribution of tweets, an average of 8 tweets per month is denoted in @DalaiLama, while @Pontifex contains an average of 48 tweets per month. However, in the case of the second religious leader, an exponential increase is displayed (+416 tweets) since the beginning of the year 2020 when + 45% of tweets were published for 7 months. On average, during the year 2019, 42 tweets/month were published, while in 2020 it was 59 tweets/month. In short, it is noted that @DalaiLama maintains a symmetry in the monthly amount of tweets in both 2019 and 2020, however, @Pontifex in addition to publishing more than @DalaiLama has also progressively increased the number of publications in 2020 (Fig. 2). Microblogging services such as Twitter are configured to use hashtags (#), this tool facilitates the search and filtering of specific information. During the 19 months reviewed, the predominance of @Pontifex in the use of symbols is clearly shown. In fact, the

283

PT

ER

RETRACTED CHAPTER: Semantic Analysis on Twitter of Two Religious’ Leaders

H

A

Fig. 2. Distribution of tweets from @Pontifex and @DalaiLama.

R

ET

R

A

C

TE

D

C

20 most repeated hashtags come from that account (Fig. 3), highlighting Catholic festivities such as #HomilySantaMarta, #Lent, #Angelus, #Easter, as well as events such as #GeneralAudience, #LaudatoSi5 and #PBC2019, while @DalaiLama only incorporates them in one of its sentences, as happens with #InternationalWomensDay; #PeaceDayChallenge; #JanAndolan; #SwachhBharatMission; #plasticfree, #InfinitePotentialFilm, #DavidBohm; #QuantumPotential.

Fig. 3. Most used hashtags 2019–2020

284

A. Torres-Toukoumidis et al.

C

H

A

PT

ER

Then, for the sentiment analysis, the modulated compilation of the phrases emitted by both religious leaders leads to demonstrate significant particularities in their messages. In the first instance, @DalaiLama expresses a notion of his doctrine through values such as compassion (52), happiness (43), strength (32), love (31), peace (29), tolerance (26) adding also general concepts such as something (27), others (27), life (24). On the other hand, @Pontifex reverberates the identity with terms properly seated in doctrine: god (31), Christian (30), catholic (29), church (26), Vatican (19), cardinal (17) and jesus (16). Similarly, it expresses actions correlated with its religious vision as charity (29), philanthrophy (25), pray (21), allow (15). Finally, the presence of contextual circumstances is uniquely evident by adding the reiteration of the term COVID (20) (Fig. 4).

TE

D

Fig. 4. Top 15 most frequent terms in @DalaiLama and @Pontifex

ET

R

A

C

Both religious leaders have different communicational strategies. A greater terminological heterogeneity of the @DalaiLama is evident, accentuating a nucleus of words related to human values, but at the same time accompanying it with words of lesser frequency that can be inferred towards an orientation of political criticism: Tibetan, Chinese, censored, security. On the other hand, @Pontifex builds its terminological core towards the identity of the religion it professes by linking it with reflections on the global situation echoing terms such as COVID, media, couple, victims. In short, the Dalai Lama and Pope Francis have developed a different lexical repertoire in the messages on their Twitter accounts, maintaining a unique interaction with their followers.

R

4 Conclusions and Discussions According to the results obtained, multiple edges are discovered that arise from the interaction exposed by both religious leaders in this microblogging platform, while at the same time new disjunctions are built for future research. The first edge is ordered towards a quantitative variable that responds to the frequency of emission of messages in Twitter, this purely numerical objective allowed observing the concurrent use of Pope Francis in the writing of 280 characters. From a descriptive point of view, curiously none of the religious leaders follow each other, in fact, the Dalai Lama does not follow any other Twitter account, while Pope Francis follows his other accounts presented

RETRACTED CHAPTER: Semantic Analysis on Twitter of Two Religious’ Leaders

285

A

C

TE

D

C

H

A

PT

ER

in other languages. Similarly, both redirect their accounts to websites and contain a defined location. Regarding @Pontifex, it links to the official Vatican news website (vaticannews.va) and of course is in Vatican City, while @DalaiLama links to its own website (dalailama.com) and is geolocated in Dharamsala, India. As for the second edge, hashtags play a heterogeneous role in coordinating online communities and search-related issues, so it can be considered as a resource that stands out in both religious leaders to indicate a post that is considered important to interpret opinion and feeling in social network discourses. On the part of Pope Francis, the hashtags emphasize events linked to the doctrine of the Church, while the Dalai Lama directs their use to current issues that are reiterated in the communication spectrum, but of a more general nature. On the other hand, the third part of the study sought to go deeper into the lexicon constructed in the tweets by means of a computerized text mining tool, managing to extract the most frequent terms, leading to the interpretation of divergent communicative strategies. @DalaiLama uses an inter-religious communicative strategy through concepts that could permeate different doctrines; meanwhile, the strategy of @Pontifex is constructed towards the promotion of institutional identity, combined with the contemporary situation. Some of the limitations presented in the study were the difficulties in systematizing the tweets and the second, maintaining the impartial scientific correspondence of the information, keeping a discourse that does not alter religious sensitivities. In short, analyzing the communicative presence of both religious leaders has generated significant findings that should be extrapolated to other researches, such as the review of other social networks and the prospective use of other emerging impact platforms such as TikTok. We also invite descriptive comparisons of other religious leaders, as well as in-depth profiles of regional and local leaders. In fact, in support of knowledge mainstreaming, the application programming interface components presented in this document are freely accessible and can be implemented in other studies related to social networks, thus producing new contributions for the academic community.

ET

R

References

R

1. Liu, M.: Dalai Lama, Twitter Rock Star: The Virtual Influence of His Holiness. https://bit.ly/ 3j4yrlH. Accessed 21 Oct 2020 2. Macale, S.: Dalai Lama and Archbishop Desmond Tutu to Hangout on Google+. https://bit. ly/3k4ynDF. Accessed 21 Oct 2020 3. Choney, S.: Big-time ‘hangout’: Dalai Lama and Desmond Tutu on Google+. https://on.today. com/3403HxN. Accessed 21 Oct 2020 4. Fung, B., Boorstein, M.: Pope Francis calls the Internet ‘a gift from God’. The Washington Post. https://wapo.st/3lTpqgW. Accessed 21 Oct 2020 5. Lichterman, P.: Religion and the construction of civic identity. Am. Sociol. Rev. 73(1), 83–104 (2008) 6. Campbell, H.: Spiritualising the internet: uncovering discourses and narratives of religious internet usage. Online Heidelberg J. Relig. Internet 1(1), 1–26 (2005) 7. Cannon, D.: Who do you say that I am? A history of how religion communicators have often avoided “public relations.” J. Public Relat. Res. 27(3), 280–296 (2015)

286

A. Torres-Toukoumidis et al.

R

ET

R

A

C

TE

D

C

H

A

PT

ER

8. Rupp, W., Smith, A.: A study of the interrelationships between the internet and religious organizations: an application of diffusion theory. Serv. Mark. Q. 24(2), 29–41 (2002) 9. Larsen, E.: Wired Churches, Wired Temples. Pew Research Center. https://pewrsr.ch/2T1 St5y. Accessed 25 Oct 2020 10. Morehouse, J.: Favorite for Jesus, Retweet for Buddha: A Content Analysis of Religious Leaders on Twitter. University of Houston Press (2015) 11. Tilson, D., Chao, Y.: Saintly campaigning: devotional-promotional communication and the U.S. tour of St. Therese’s relics. J. Media Relig. 1(2), 81–104 (2002) 12. Greer, C.F., Ferguson, D.A.: Using Twitter for promotion and branding: a content analysis of local television Twitter sites. J. Broadcast. Electron. Media 55(2), 198–214 (2011) 13. Aiello, L., Petkos, G., Martin, C., Corney, D., Papadopoulos, S., Skraba, R., Goker, A.: Sensing trending topics in Twitter. IEEE Trans. Multimedia 15(6), 1268–1282 (2013) 14. Duggan, M., Ellison, N., Lampe, C., Lenhart, A., Madden, M.: Demographics of Key Social Networking Platforms. Pew Internet Research. https://pewrsr.ch/3511QIC. Accessed 26 Oct 2020 15. Vargo, C., Guo, L., McCombs, M., Shaw, L.: Network issue agendas on twitter during the 2012 U.S. presidential election. J. Commun. 64(2), 296–316 (2014) 16. Kim, H., Jang, M., Kim, S., Anan, W.: Evaluating sampling methods for content analysis of Twitter data. Soc. Media+ Soc. 4(2), 120–135 (2018) 17. Toth, T., Farkas, J.: The terminologies of two religious leaders rhetoric about communities in pope Francis’and Dalai Lama’s tweets. Eur. J. Sci. Theol. 15(5), 159–178 (2019)

Step Towards Predicting Patient Length of Stay in Intensive Care Units Cristiana Neto1 , Gabriel Pontes1 , Alexandru Domente2 , Francisco Reinolds2 , Jos´e Costa2 , Diana Ferreira1 , and Jos´e Machado1(B) 1

Algoritmi Research Center, University of Minho, Campus of Gualtar, Braga, Portugal {cristiana.neto,gabriel.pontes,diana.ferreira}@algoritmi.uminho.pt [email protected] 2 University of Minho, Campus of Gualtar, Braga, Portugal {pg41063,a82982,a82136}@alunos.uminho.pt

Abstract. Increasingly, hospitals are collecting huge amounts of data through new storage methods. These data can be use to extract hidden knowledge, which can be crucial to estimate the length of stay of admitted patients in order to improve the management of hospital resources. Hence, this article portrays the performance analysis of different data mining techniques through the application of learning algorithms in order to predict patients’ length of stay when admitted to an Intensive Care Unit. The data used in this study contains about 60,000 records and 28 features with personal and medical information. A full analysis of the results obtained with different Machine Learning algorithms showed that the model trained with the Gradient Boosted Trees algorithm and using only the features that were strongly correlated to the patient’s length of stay, achieved the best performance with 99,19% of accuracy. In this sense, an accurate understanding of the factors associated with the length of stay in intensive care units was achieved. Keywords: Length of stay · Hospitalization stay unit · Data mining · Classification · CRISP-DM

1

· Intensive care

Introduction

Hospitals have been benefiting from a technology revolution, particularly through the increasing adoption of health information systems [1]. With the enormous growth of data and the complexity of the stored information, processing and analysing it to improve hospital management becomes difficult but very important. Data Mining (DM) is a powerful tool that can help healthcare institutions to extract useful knowledge from a great amount of data [2,3]. Since hospitals are faced with very limited resources including beds to receive patients, specially during this time of the COVID-19 pandemic, the Length of c The Author(s), under exclusive license to Springer Nature Switzerland AG 2021  ´ Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 287–297, 2021. A. https://doi.org/10.1007/978-3-030-72654-6_28

288

C. Neto et al.

Stay (LOS) is a crucial factor for a better planning and management of hospital resources. In this context, the prediction of a patient’s LOS will help hospitals achieve many goals, such as better benchmark performance in terms of profitability and patient care efficiency, due to the fact that LOS is an essential measure of healthcare utilization and a determining factor in hospitalization costs [4,5]. Several studies have focused on predicting the LOS in ICU [5,6]. Although, the results observed in these studies showed an opportunity for improvement. Thus, this article will focus on the application of DM techniques to define a model capable of predicting the LOS of patients admitted to an Intensive Care Unit (ICU). All the results obtained are going to be interpreted taking into consideration the impacts on medical and management purposes. This article is organized in four sections. The following section describes the problem and the methodology used to address it. The next section details every phase of the chosen DM methodology. The third section contains the analysis and discussion of the results. Finally, the last section highlights the contributions of this research and potential future work.

2

Methodologies, Materials and Methods

The chosen methodology to conduct this DM process was CRISP-DM (Cross Industry Standard Process for Data Mining) which contemplates six distinguished and flexible phases: Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation and Deployment [7]. It is also imperative to support this study by using technologies whose features guarantee a effective execution in each phase of the established DM methodology. Thus, the KNIME data analytic platform was used in order to transform and modify the data (Data Preparation). Unlike other DM tools, KNIME holds a wide range of simple and optimized data transformation features. On the other hand, the development of the remaining CRISP-DM phases was made on the RapidMiner software, a data platform that provides an integrated environment for Machine Learning (ML) and Predictive Analytics, being widely used in DM due to its high usability compared with other alternatives. So, after data preparation occurs on KNIME, its output data is allocated as input on RapidMiner, which is then used to model a set of ML algorithms and to individually assess each one’s performance. Thus, this study used different supervised algorithms which can be divided in three major groups: tree-based ML algorithms, namely Decision Tree (DT), Random Forest (RF) and Gradient Boosted Trees (GBT), neural network-based ML algorithms, namely Neural Net (NN) and Deep Learning (DL) and, lastly, two more algorithms: k-Nearest Neighbour (k-NN) and Na¨ıve Bayes (NB). 2.1

Business Understanding

At this stage, the understanding of the objectives and the requirements from the business’ point of view is of utmost importance.

Step Towards Predicting Patient Length of Stay in Intensive Care Units

289

In this work, the focus resides on the problem of scheduling hospital admissions, where the greatest difficulty is maintaining an adequate number of beds for patients who need intensive care and reducing the high costs that the hospital presents due to a poor management of the resources. In order to provide ICUs with valuable information, this study will include several aspects, such as the detection of the highest correlated features to the LOS, the determination of the optimal parameter values of the given ML algorithm and the prediction of the LOS by constructing different DM models (DMM). Afterwards, it is going be exposed a one-to-one comparative analysis for each DMM, with or without optimized parameters and strongly correlated features. In this work, different techniques will be compared in terms of their performance and the results obtained for the different metrics selected in order to find the approach that best suited the problem and the data used. 2.2

Data Understanding

In this phase, the analysis of the collected data is performed, as well as the identification of the relevant attributes, so that a better perception of the existing data can be obtained. The used dataset contains about 60,000 records from a clinical computer system that was used between 2001 and 2008 on Beth Israel Deaconess Medical Center, a medical care unit located in Boston. Most of the stored attributes are essential for determining the length of hospitalization in the ICU of this hospital, including patients’ details as well as reasons for admission, discharge details, laboratory test results, and other medical features [8]. Each of the records is described by a total of 28 attributes as presented in Table 1. Through a more detailed analysis of Table 1, the similarity between the LOSdays and LOSgroupNum features stands out. While the first one records the number of days each patient stayed in the ICU, the second one discretizes the LOSdays attribute in four bins, giving a four-day range to each category, except for the last label (as seen on the table footnote). As these two attributes are completely related, which of these features should be abandoned will be discussed during the next phase of this DM process, so that any inconsistencies in the models can be avoided. In this phase, it was also noticed that all numerical features needed some correction regarding inconsistencies on high standard deviation values relatively to the feature’s average value. 2.3

Data Preparation

The first step on the data preparation was to ensure that raw data did not contained any duplicate, inconsistent or incomplete information. After a quick overview, neither of these situations were verified on the used data. Next, all nominal personal information was transformed into numerical features with the purpose of allowing the use of ML algorithms like Neural Net, that only runs over numerical features.

290

C. Neto et al. Table 1. Dataset attributes description

Attribute

Description

Attribute

Description

hadm id

Hospital admission identifier

NumOutput

Number of outputs performed

gender

Patient’s gender

NumCPTevents

Number of performed current procedural terminology (CPT) events

age

Patient’s age (in years)

NumInput

Number of performed inputs

insurance

Patient’s insurance type

NumLabs

Number of performed laboratory tests

religion

Patient’s religion

NumMicroLabs

Number of performed microbiology tests

marital status

Status Patient’s Marital status

NumNotes

Number of written notes and summaries

ethnicity

Patient’s ethnicity

NumRx

Number of performed x-rays

admit type

Admission type

NumProcEvents

Number of performed procedure events

admit location

Admission location

NumTransfers

Number of performed transfers

AdmitDiagnosis Diagnosis Admission Diagnosis

NumChartEvents Number of performed chart events

NumDiagnosis

Number of diagnoses entered to the system

ExpiredHospital

NumCallouts

Number of performed callouts

TotalNumInteract Total number of performed interactions

NumProcs

Number of performed procedures

LOSdays

Indicates if the patient has died or not

Length of Stay in Intensive Care Unit

AdmitProcedure Performed procedure LOSgroupNum Length of Stay in on admission Intensive Care Unita a 0: LOS 0-4 days; 1: LOS 4-8 days; 2: LOS 8-12 days; 3: LOS > 12 days

Thenceforth, a search to detect and manage potential outliers among the data was performed using a simple interquartile range algorithm capable of detecting and later removing outliers. This algorithm relies on data’s interquartile range value, measured with Eq. 1, where Q1 (first quartile) and Q3 (third quartile) represent, respectively, a quarter and three quarters of all data records [9]. IQR = Q3 − Q1

(1)

However, to detect outliers, a few more steps must be taken. After calculating the IQR values, each one must be multiplied by a constant named Interquartile Range Multiplier (IQRM). This constant’s correct value is 1.5 in order to discern outliers, as shown in Eq. 2. C = IQR ∗ 1.5 (2)

Step Towards Predicting Patient Length of Stay in Intensive Care Units

291

Lastly, the C is subtracted to Q1 and added to Q3, establishing a range of non-outlier records. So, any record located outside this range is labelled as an outlier, as represented in Eq. 3 and Eq. 4. x < Q1 − C → Outlier

(3)

x > Q3 + C → Outlier

(4)

After detecting and removing all outliers, the continuous features of the dataset were discretized so that a better performance is attained during the modelling phase. In what regards the discretization, only the age feature was affected, applying a binning process with equivalent ranges of 10 years. Then it was necessary to equip the data with more valuable features to the forecast of a patient’s LOS in an ICU. To do so, two new medical information features were generated, representing respectively the number of diagnosis and procedures made to a patient per internment day. These two features were calculated based on Eq. 5 and Eq. 6, respectively. N umP rocs LOSdays

(5)

N umDiagnosis LOSdays

(6)

N umP rocsP erDay =

N umDiagsP erDay =

In an opposite direction, but with the same purpose of improving the model’s performance, an overview of every feature was made to understand each one’s relevance. During this analysis, it was found that ExpiredHospital, a binary feature that indicates if a patient died or not in the ICU, only presents records of patients who have not died (value = 0). So, in this situation, the approach was to remove this feature due to its lack of ability to influence the DMM. On the other hand, the aforementioned LOSdays feature indicates the number of days a patient has spent in the ICU, whilst the feature LOSgroupNum gives the same exact information with the only difference of being discretized. Therefore, it is easy to understand that these two attributes are strongly correlated and the use of one in order to forecast another might bias the developed models and, in a worst-case scenario, inducing the model to return corrupted and invalid results. So, to avoid the preceding described complications, the LOSdays attribute was removed from the data. Another removed feature was hadm id since it represents an identifier and does not add predicting value to the DMM. Lastly, every feature was normalised into a value range between 0 and 1, using RapidMiner in order to have the possibility of denormalizing the features when the model’s execution is complete, allowing a better data analysis. 2.4

Modeling

After the data preparation, we are able to begin the construction of the DMM and, in the next phase, to compare different their performances.

292

C. Neto et al.

The first step of this phase was to acknowledge which features were more correlated to the target attribute using the Weight by Correlation in RapidMiner. Two of the features – NumTransfers and NumDiagsPerDay – were evaluated with a strong negative correlation, with values of −0.71 and −0.69, respectively. From this point forth, the DMM will be compared with and without these two features, in order to study its influence in the forecast. Ultimately, each ML algorithm has many parameters for its execution, and the results may vary with a different set of values assigned to each parameter field. Therefore, a model tuning is paramount, before the model’s execution, using the RapidMiner Optimize Parameters operator for each algorithm. Along with the optimized parameters being fed to each algorithm, it is also beneficial to execute the models with and without the strong correlated features. Therefore, a total of four scenarios were created: S1 - Without correlated attributes and the default parameters; S2 - With correlated attributes and the default parameters; S3 - Without correlated attributes and the optimized parameters; S4 - With correlated attributes and the optimized parameters. Cross Validation with 10 folds was the sampling method chosen. Taking into account that seven ML algorithms were selected to be executed during the modeling phase, it is simple to deduce that a total of 28 models were developed. 2.5

Evaluation

During this phase, the DMM were evaluated using several statistical metrics in order to provide a quality assessment of the model. Thus, the confusion matrix was analized, showing in a tabular structure the True Negatives (TN), the True Positives (TP), the False Negatives (FN) and the False Positives (FP). As a multiclass prediction – the LOSgroupNum feature have four distinct possible outcomes –, the confusion matrix diagonal exposes all the TP entries referred to every successful forecast, while each result’s line and column exhibit the FN and FP entries, respectively. In our case of study, these two are worrisome fields when taking into account the fact that both represent wrong forecasts of the target feature. Lastly, the other values in the confusion matrix refer to TN records or, in other words, instances correctly labelled as having a different length of stay. Through confusion matrix values, it’s possible to estimate many other performance metrics. First, the model’s accuracy is estimated through Eq. 7 [2]. Accuracy(%) =

TP + TN ∗ 100 TP + TN + FP + FN

(7)

With this metric correctly estimated, it is very simple to deduce the classification error, that refers to the accuracy’s complementary (1 - Accuracy). Precision and Sensitivity (or Recall) metrics were also evaluated. While Precision tries to measure the proportion of positive identifications that were correct, Recall indicates the proportion of actual positives correctly identified. These two metrics are represented in Eq. 8 and Eq. 9 [2]. P recision(%) =

TP ∗ 100 TP + FP

(8)

Step Towards Predicting Patient Length of Stay in Intensive Care Units

293

TP ∗ 100 (9) TP + FN Outdistancing from the metrics derived from the confusion matrix, other metrics were evaluated. The kappa coefficient is one of them and it’s useful to multi-class and imbalanced problems or when measures like accuracy, precision or recall are not able to provide a complete performance overview of a classifier. Its values vary between 0 and 1. In addition, the RMSE (Root-mean-square error) was also used to evaluate the models. Although this measure is more common and useful on regression models, which is not the case, it measures the standard deviation of the predictions from the ground-truth, so this measure is taken as well. This metric is presented in Eq. 10, where N is the number of elements on our dataset and Oi represents the observed value whereas ei shows the actual value of the entry [10].  N  (Oi − Ai )2 (10) RM SE =  N i=1 Recall(%) =

3

Results and Discussion

After an extended overview of the results present on Table 2, some deductions become possible. Starting with the observation of each scenario, major improvements are acknowledged from models executed over algorithms with optimized parameters (S3 and S4) while worse results were recorded with default parameters (S1 and S2), which reveals the importance of an optimal running environment for each one of the used ML algorithms. So, in order to continue this topic’s analysis, for each algorithm, only the last two columns of Table 2 will be minutely analysed from now on. Paying closer attention to Table 2, an uplift from S3 to S4 stands out, which can be attributed to the two features detected as strongly correlated to the prediction attribute. In other words, the inclusion of both strongly correlated attributes in S4, allows a model to improve its results as expected, resulting in a 5% accuracy improvement. In order to understand the real behaviour of the created models, it is important to analyse other metrics. For instance, a 90% accuracy would not be “real” if 90% of the cases would pertain to only one class and an hypothetical model could only predict that class. However, the obtained results for Sensitivity and Recall reveal a striking similarity to all the Accuracy values recorded with tiny variances, allowing us to verify the Accuracy’s veracity. Lastly, for the remaining measures, it is worth noting that both Kappa Coefficient and RMSE values followed Accuracy behaviour, respectively. While the first one assigns higher coefficients to models using both optimized parameters on their algorithm and the two correlated attributes – NumTransfers and NumDiagsPerDay –, the other one reduces its error coefficient when evaluating this same kind of models.

294

C. Neto et al. Table 2. Algorithm’s performance in several categories Category

ML algorithm

S1

S2

S3

S4

Accuracy (%) Na¨ıve Bayes k-NN Neural Net Deep Learning Decision Tree Random Forest Gradient Boosted Trees

62,44 55,50 91,46 90,10 70,94 67,42 92,07

67,39 59,15 93,59 95,60 72,20 75,07 96,85

69,10 57,08 91,46 92,45 87,90 90,90 97,47

78,69 61,54 95,09 96,87 96,82 94,42 99,19

Precision (%) Na¨ıve Bayes k-NN Neural Net Deep Learning Decision Tree Random Forest Gradient Boosted Trees

62,65 57,30 91,57 90,42 75,37 74,68 92,69

68,34 61,31 93,75 95,85 76,38 81,18 96,92

82,01 61,58 91,68 92,31 87,37 91,50 97,53

85,22 65,98 95,06 96,60 96,62 94,80 99,25

Recall (%)

Na¨ıve Bayes k-NN Neural Net Deep Learning Decision Tree Random Forest Gradient Boosted Trees

61,23 51,87 90,43 88,71 68,17 61,80 91,10

67,30 56,00 93,54 94,96 68,85 70,85 96,32

56,52 49,83 90,18 91,57 87,08 89,58 97,13

70,54 55,11 94,74 96,79 96,63 93,61 99,07

Kappa

Na¨ıve Bayes k-NN Neural Net Deep Learning Decision Tree Random Forest Gradient Boosted Trees

0,466 0,355 0,878 0,858 0,579 0,507 0,886

0,540 0,410 0,909 0,937 0,594 0,630 0,955

0,518 0,362 0,878 0,892 0,828 0,869 0,964

0,680 0,432 0,930 0,956 0,955 0,920 0,988

RMSE

Na¨ıve Bayes k-NN Neural Net Deep Learning Decision Tree Random Forest Gradient Boosted Trees

0,538 0,589 0,259 0,268 0,502 0,572 0,368

0,492 0,560 0,214 0,179 0,483 0,508 0,298

0,512 0,592 0,257 0,239 0,336 0,317 0,147

0,420 0,564 0,195 0,154 0,172 0,254 0,078

Step Towards Predicting Patient Length of Stay in Intensive Care Units

295

From this point forth, another approach is taken, comparing each algorithm’s performance. Starting with k-NN, the results prove its ineffectiveness in dealing with our classification problem. It is considered to be a sensitive algorithm, which means the results can be impacted by noise or irrelevant data and as a sampling was not developed in our data, its lackluster performance is understandable on this algorithm, due to its aforementioned shortcomings. Although NB achieved better results than k-NN, it also was incapable of obtain high accuracy, precision or recall percentages. Despite being a swift and simple classifier, its assumption that every attribute is independent, thus failing to establish relations between them, results in it having a subpar performance. Analysing the two neural network-based algorithms (NN and DL), it is observed that these two live up to the expectations as a powerful, complex and high computational effort type approach, as the results shows a 90+ percentage of Accuracy in every data set executed. One can’t disregard that, for most scenarios, a small gain on the DL classifier is observed which can represent a tiebreaker on an hypothetical decision regarding the use of a NN algorithm. Proceeding to the analysis of each one of the three tree-based algorithms implemented, among which lies the best model achieved. The first one, DT, uses a single decision tree to support its decisions whereas RF uses sequentially built multiple trees. Taking this into consideration, one could erroneously come to a conclusion that RF would achieve better results, while the results proved otherwise. This outcome proves the great overall performance returned by DT, and its better usability over RF for this specific context. Finally, GBT, a RF variation capable of building its multiple decision trees at once, became our model’s benchmark. With a 99+% Accuracy, Precision and Recall allied to an emphasized small RMSE comparatively to the rest, this is the ML algorithm selected to model this DM process. Although the ability of building its decision trees increases the algorithm complexity and computational effort concurrently, when compared to a neural network algorithm such as DL, the costs are similar generating however better results for the cost it poses.

4

Conclusion

Hospitals institutions face daily challenges that come with very limited resources, specially in this COVID-19 pandemic time, and LOS is a crucial factor for better planning and management. In this context, the prediction of a patient’s LOS can help a hospital to achieve many goals such as better benchmark performance in terms of profitability, patient care and process efficiency. With that purpose, this study tried to illustrate every stage behind CRISP-DM methodology, clarifying all the steps that helped the achievement of satisfactory results in predicting how long will a patient stay in an ICU. With this study, it was also proved the success of DMM in attaining this classification goal. Overall a great result was obtained using the GBT algorithm with optimized parameters and with the presence of all attributes, achieving an accuracy of 99,19%, a precision of 99,25% and a recall of 99,07%, percentages high enough to prove that the main goal was successfully accomplished.

296

C. Neto et al.

This study proves that, for the used dataset, the GBT algorithm is capable of predicting LOS in ICU with an accuracy of approximately 99%. However, this result does not mean that the model would be able to predict LOS in ICU with the same accuracy when using other data, being required a larger volume of data to assess the model’s generalization ability. In this sense, it should be noted that the process described in this study lacks a few requirements that reveal potential in improving the obtained results and, consequently, achieving more rigorous conclusions through valuable information. Firstly, during the data preparation stage, a step of sampling could have been developed in order to balance the dataset. For instance, with a SMOTE sampling technique capable of generating new records of the minority class. With this, algorithms such as k-NN would see a big increase in performance due to its better results over balanced data. Secondly, the data extraction from the hospital could be improved in order to record patient’s individual data concerning previous entries in ICU and whether those occurred in recent years, especially at this time of the COVID-19 pandemic, where the collection of data related to this context would be of high importance. Although being relevant to conduct this study with data from the ICU during COVID-19, the acquisition of the data is still challenging. With a recent and more detailed raw data, a more precise model could be developed. Acknowledgements. This work has been supported by FCT – Funda¸ca ˜o para a Ciˆencia e Tecnologia within the R&D Units Project Scope: UIDB/00319/2020.

References 1. Lee, T., Liu, C., Kuo, Y., Mills, M., Fong, J., Hung, C.: Application of data mining to the identification of critical factors in patient falls using a web-based reporting system. Int. J. Med. Inform. 80, 141–150 (2011) 2. Neto, C., Brito, M., Lopes, V., Peixoto, H., Abelha, A., Machado, J.: Application of data mining for the prediction of mortality and occurrence of complications for gastric cancer patients. Entropy 21(12), 1163 (2019) 3. Han, J., Pei, J., Kamber, M.: Data Mining: Concepts and Techniques, 3rd edn. Elsevier, Amsterdam (2011) 4. Baek, H., Cho, M., Kim, S., Hwang, H., Song, M., Yoo, S.: Analysis of length of hospital stay using electronic health records: a statistical and data mining approach. PLoS ONE 13(4), e0195901 (2018) 5. Veloso, R., et al.: Real-time data mining models for predicting length of stay in intensive care units. In: KMIS 2014 - Proceedings of the International Conference on Knowledge Management and Information Sharing, pp. 245–254 (2014) 6. Navaz, A.N., Mohammed, E., Serhani, M.A., Zaki, N.: The use of data mining techniques to predict mortality and length of stay in an ICU. In: 12th International Conference on Innovations in Information Technology (IIT), pp. 1–5. IEEE (2016) 7. Neto, C., Peixoto, H., Abelha, V., Abelha, A., Machado, J.: Knowledge discovery from surgical waiting lists. Procedia Comput. Sci. 121, 1104–1111 (2017) 8. Scarlat, A.: MIMIC3d aggregated data - ICU aggregated data as number of interactions between patient and hospital, Version 1. https://www.kaggle.com/drscarlat/ mimic3d

Step Towards Predicting Patient Length of Stay in Intensive Care Units

297

9. Whaley, D.L.: The Interquartile Range: Theory and Estimation. Electronic theses and dissertations. Paper 1030 (2005) 10. Martins, B., Ferreira, D., Neto, C., Abelha, A., Machado, J.: Data mining for cardiovascular disease prediction. J. Med. Syst. 45(1), 1–8 (2021)

Numerical Validation of Half Precision Simulations Fabienne J´ez´equel1,2(B) , Sara sadat Hoseininasab1 , and Thibault Hilaire1 1

2

Sorbonne Universit´e, CNRS, Laboratoire d’Informatique de Paris 6, LIP6, 75005 Paris, France {Fabienne.Jezequel,Sara.Hoseininasab,Thibault.Hilaire}@lip6.fr Universit´e Panth´eon-Assas, 12 place du Panth´eon, 75231 Paris CEDEX 05, France

Abstract. In this article, we show how to control the numerical quality of half precision computations using stochastic arithmetic. The CADNA library that is used to estimate rounding errors and detect numerical instabilities in floating-point codes has been improved and extended to support half precision. A performance gain up to a factor 20 has been observed w.r.t. accuracy estimation in arbitrary precision. Furthermore we present how to generate codes mixing half, single, and double precision with a requested accuracy on results providing a user-defined numerical quality of the code. Control of robustness and floating-point auto-tuning taking into account half precision have been successfully performed on various numerical simulations, in particular a control application. Keywords: Numerical quality · Numerical validation · Rounding errors · Auto-tuning · Discrete stochastic arithmetic · Floating-point arithmetic · Half precision · Mixed precision

1

Introduction

Among the different tests that must be performed to validate a code, numerical validation is crucial because rounding errors generated by its floating-point operations can invalidate its results. Indeed controlling the robustness of a simulation code includes the analysis of its numerical quality. Numerical simulations tend to be carried out in reduced precision (e.g. single or half precision instead of double precision) or in mixed precision. Since reducing the precision has an impact on performance, memory usage, and energy consumption, using half precision can be particularly advantageous [1,5]. However, because of the low mantissa-length and the limited range of that format, the accuracy of half precision computation should be carefully controlled in order to ensure code quality. We respond to this challenging issue with the following contributions. – We show how to estimate rounding errors and detect numerical instabilities in any half precision computation. The related software development has been integrated in the CADNA library [2]. c The Author(s), under exclusive license to Springer Nature Switzerland AG 2021  ´ Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 298–307, 2021. A. https://doi.org/10.1007/978-3-030-72654-6_29

Numerical Validation of Half Precision Simulations

299

– We have extended the PROMISE software [4] which performs floating-point auto-tuning such that it can automatically generate codes mixing half, single and double precision. – Thanks to the new version of CADNA, the numerical quality of various simulations in half precision has been analysed, in particular, a control application and a Conjugate Gradient solver. Furthermore versions of these codes in mixed precision (including half precision) have been provided by the new version of PROMISE. The remaining part of this paper is organized as follows. Section 2 reminds the principles of the rounding error estimation using the Discrete Stochastic Arithmetic and the floating-point auto-tuning. Section 3 shows how floatingpoint computation can be carried out in half precision. Sections 4 and 5 detail the extension of the accuracy estimation and auto-tuning to half precision arithmetic. Finally, Sect. 6 presents some numerical experiments, and conclusions are given in Sect. 7.

2

Control of Numerical Quality and Floating-Point Auto-tuning

2.1

Principles of DSA

Discrete Stochastic Arithmetic (DSA) [9] is an automatic method for rounding error analysis based on a probabilistic approach. DSA can provide an estimation of the numerical quality of results, especially in large scale applications. DSA allows one to estimate the number of correct digits in computed results by executing the user programs several times using a random rounding mode: either rounding to +∞ or to −∞ with the same probability. Therefore, the computer’s deterministic arithmetic is replaced by a stochastic arithmetic, where each arithmetic operation is performed N times before the next one is executed. DSA supplies us with N samples R1 , · · · , RN of each computed result R and the number of correct digits in R is estimated using a statistical test. 2.2

Accuracy Estimation by CADNA and SAM

DSA is implemented, on the one hand, in the CADNA1 library [2] that can be used to control the accuracy of programs in single, double and/or quadruple precision, and, on the other hand, in the SAM2 library [3] that estimates rounding errors in arbitrary precision programs. Thanks to three executions of the user program with the random rounding mode, CADNA and SAM estimate, with the probability 95%, the number of correct digits of any computed result. Their codes are based on new numerical types: the stochastic types. Each stochastic variable contains three values of the corresponding numerical type and an integer 1 2

http://cadna.lip6.fr. http://www-pequan.lip6.fr/∼jezequel/SAM.

300

F. J´ez´equel et al.

to store its number of correct digits. Arithmetic operators, comparison operators, all the mathematical functions are overloaded for these stochastic types. Therefore the use of CADNA or SAM in a program requires only a few modifications: essentially changes in the declarations of variables and in input/output statements. CADNA and SAM can detect numerical instabilities which occur during the execution of the code. Such instabilities are usually generated by numerical noise, i.e. a result having no correct digits. 2.3

Floating-Point Auto-tuning: The PROMISE Software

The PROMISE3 tool [4], based on CADNA, aims at reducing in numerical programs the number of double precision variable declarations in favor of single precision ones. From an original program and a requested accuracy on the result, PROMISE provides a transformed program having a maximum number of variables declared with a lower precision and computing a result that satisfies the accuracy constraint. The search for a suitable type configuration is performed with a reasonable complexity thanks to the Delta Debug algorithm [10] based on a hypothesis-trial-result loop.

3

Half Precision Computation sign

exponent (5 bits)

mantissa (10 bits)

Fig. 1. binary16 format

Half precision is referred to as binary16 in the IEEE 754-2008 standard [6]. As shown in Fig. 1, a binary16 floating-point number consists of a sign bit, a 5-bit long exponent, and a 10-bit long mantissa. Although 10 bits are explicitly stored to represent the mantissa, its precision is actually 11 bits. Because log10 (211 ) ≈ 3.3, the accuracy of a binary16 number is at most 3 decimal digits. In binary16, the minimum strictly positive (subnormal) value is 2−24 ≈ 5.9610−8 and the maximum representable value is 65504. The binary16 format is increasingly supported by hardware. It is available for instance on ARM CPUs from version 8.2, on NVIDIA Pascal and Volta GPUs, on AMD Radeon Instinct MI25 GPU. binary16 computation can also be emulated, for instance using the FlexFloat4 library [8] that supports multiple floating-point formats or the IEEE 754-based half-precision floating-point library developed by C. Rau5 . In this paper, we focus on the binary16 format, however, another 16-bit floating-point format called bfloat16 is available, for instance on Google Cloud TPUs and ARM 3 4 5

http://promise.lip6.fr. https://github.com/oprecomp/flexfloat. http://half.sourceforge.net.

Numerical Validation of Half Precision Simulations

301

NEON CPUs. Compared to binary16 numbers, bfloat16 numbers benefit from a wider range thanks to their 8 bit-exponent, but have a lower significand precision (8 bits instead of 11).

4 4.1

Controlling the Numerical Quality of Half Precision Computation Extension of CADNA to Half Precision

CADNA has been extended to control the robustness of codes with half precision declarations. On processors that do not support half precision, CADNA relies on the half precision floating-point library developed by C. Rau already mentioned in Sect. 3. This half precision library has been improved to enable computation mixing of half, single and double precision. CADNA has also been extended to be used on the ARM v8.2 processor that supports half precision computation. A new stochastic type, half st, associated with emulated or native half precision has been introduced in CADNA. A half st variable consists in three half precision floating-point values and an integer to store the accuracy. Relational and arithmetic operations, mathematics and intrinsic functions have been redefined for this new type. With CADNA, for performance reasons, the rounding mode is changed in an implicit way [2] as – a ⊕+∞ b = − (−a ⊕−∞ −b) (similarly for ) – a ⊗+∞ b = − (a ⊗−∞ −b) (similarly for ) where ⊕+∞ and ⊗+∞ (resp. ⊕−∞ and ⊗−∞ ) are the floating-point operations rounded towards +∞ (resp. −∞). The results of each rounding mode can be obtained from computation made in the other rounding mode. With native and emulated half precision, operations involving half st variables are performed with the random rounding mode thanks to the aforementioned arithmetic properties. The rounding mode is set once to rounding to +∞ in a CADNA initialization function, such that operations involving stochastic variables based on native types can be performed with the random rounding mode, without any explicit change of the rounding mode. In the library developed by C. Rau, half precision operations can be performed with any rounding mode of the IEEE standard. The rounding mode, which is by default rounding to nearest, has been set to rounding to +∞ thanks to a parameter change. Furthermore, because of the limited range of half precision numbers, the detection of new kinds of instabilities can been enabled: overflow and underflow caused by half precision computation. Codes including half precision can be controlled by this new CADNA version using clang++ or g++ on processors supporting half precision or thanks to emulated half precision. 4.2

Performance Tests

Performance tests have been carried out using g++ 7.4.0 on a 2.5 GHz Intel Core i7 processor and using g++ 7.5.0 on a 2.26 GHz ARM v8.2 processor, both

302

F. J´ez´equel et al.

with 4 MB cache memory. The latter is used in an Nvidia Jetson AGX Xavier module designed for embedded computation. Figures 2 and 3 present execution times for matrix multiplication on Intel Core i7 processor where half precision is emulated and on ARM v8.2 processor using native half precision. With CADNA computation has been carried out with no instability detection, and also with the detection of all kinds of numerical instabilities. Our aim here is to show the feasibility of numerical validation in half precision and to compare the cost of CADNA in half, single and double precision. The matrix multiplication is performed on one core using a non-optimized algorithm based on 3 nested loops.

Fig. 2. Performance of the multiplication of matrices of size 500

Fig. 3. Performance of the multiplication of matrices of size 2000

In the CADNA codes, a matrix is a stochastic array. The matrix values are accessed from a structure that implies indirect memory access. With CADNA, accessing matrix values is expensive, especially for small problems. Therefore as the matrix size increases, the cost of CADNA w.r.t. classic computation decreases. The higher execution times and the higher cost of CADNA on ARM v8.2 processor than on Intel Core i7 processor can be explained by the higher cost of memory access. One can observe that the cost of CADNA is not particularly higher in half precision than in single or double precision. The cost of

Numerical Validation of Half Precision Simulations

303

CADNA in half precision w.r.t single or double precision is even significantly lower when small matrices are multiplied on Intel Core i7 processor. In that case, because of half precision emulation, the ratio computation/memory access is more favorable to CADNA in half precision than in other precisions. The detection of all kinds of numerical instabilities mainly requires tests and accuracy estimation, no extra access to matrix values. Its cost w.r.t. no instability detection is about 2 on Intel Core i7 processor and 1.4 on ARM v8.2 processor. Furthermore, it has been observed that the cost of underflow/overflow detection in half precision is a factor between 1.3 and 1.6 on Intel Core i7 processor and approximately 1.2 on ARM v8.2 processor. The performance of CADNA and SAM for the numerical validation of half precision codes have been compared on Intel Core i7 processor where CADNA uses emulated half precision. For different matrix sizes, Fig. 4 presents the execution times of matrix multiplication computed with CADNA in half precision and with SAM using 11-bit mantissa length variables. A log scale is used for the y-axis to improve the readability of the results obtained. Two instability detection levels have been chosen: no instability detection and the detection of all kinds of numerical instabilities.

Fig. 4. Execution time of matrix multiplications in half precision with CADNA and SAM

One can observe that the cost of SAM w.r.t. CADNA is a factor between 13 and 20. Both with CADNA and SAM the detection of all instabilities costs about 2 w.r.t. no instability detection. As a remark, CADNA has an extra feature compared to SAM, it enables one to detect underflows and overflows in operations involving half precision variables.

5

Floating-Point Auto-tuning Using Half Precision

A new version of PROMISE has been developed in order to generate mixed precision codes including half precision. This new version also benefits from various

304

F. J´ez´equel et al.

improvements. It is more user-friendly and has better performance. Indeed the search for a valid type configuration can be accelerated thanks to information provided by the user, such as variables that should have the same type.

initial code

Delta Debug

instrumented code

Delta Debug

double

double

float

comparison double

CADNA

float

comparison double

step 1

float

half

mixed-precision code

step 2

PROMISE

reference

Fig. 5. PROMISE dataflow

PROMISE is based on the Delta-Debug algorithm [10], initially proposed for debugging programs using a scientific approach of hypothesis-trial-result loop. The approach is here used to test the numerical quality of the code with various precisions. From a C/C++ code having a set of variables to be considered (with simply changing their type to a generic PROMISE-adapted type), PROMISE uses Discrete Stochastic Arithmetic (with CADNA) to determine a reference result. This reference is then used to verify the numerical quality of subsequent results obtained with several mixed-precision codes. The search for a valid type configuration relies on the Delta-Debug algorithm, as shown in Fig. 5. If PROMISE is run in half, single and double precision, the Delta Debug algorithm is executed twice. First, PROMISE determines which variables should stay in double and which ones can be relaxed to single precision. Then, from the single precision variables, PROMISE determines which ones can be relaxed to half precision.

6

Numerical Experiments

The numerical experiments described in this section have been carried out using g++ 7.4.0 on a 2.5 GHz Intel Core i7 processor with 4 MB cache memory where half precision is emulated. 6.1

Determinant of Hilbert’s Matrix

1 for i = The determinant of Hilbert’s matrix of size n defined by Hi,j = i+j−1 1, · · · , n, j = 1, · · · , n is computed using Gaussian elimination without pivoting. n The determinant D is the product of the computed pivots: D = i=1 pi . Table 1

Numerical Validation of Half Precision Simulations

305

Table 1. Determinant of Hilbert’s matrix of size 3 with and without CADNA in half and double precision Without CADNA Double

Half

With CADNA Double

Half

p1 1.00000000000000e+00 1.000000e+00 1.00000000000000e+00 1.00e+00 p2 8.33333333333333e−02 8.325195e−02

8.33333333333333e−02 8.3e−02

p3 5.55555555555553e−03 5.310059e−03

5.5555555555555e−03

5.e−03

D 4.62962962962960e−04 4.420280e−04

4.6296296296296e−04

4.e−04

presents the results obtained with and without CADNA in double and half precision for Hilbert’s matrix of size 3. The digits in common with the exact results are displayed in bold. Without CADNA, results computed in half precision are casted to single precision to be printed. CADNA displays only the digits not affected by rounding errors: 14 or 15 digits in double precision and 1 to 3 digits in half precision. No numerical instability, nor underflow, nor overflow is detected. The digits displayed by CADNA in double and half precision are those of the exact results. 6.2

Control Application

This example is a multi-input, multi-output controller from the automotive context. It is part of an active controller of vehicle longitudinal oscillations [7]. It is a linear State-Space system of the form  x(k + 1) = Ax(k) + Bu(k) y(k) = Cx(k) + Du(k) where u(k) ∈ R5 is the vector of inputs at time k, y(k) ∈ R7 the vector of outputs at time k and x(k) ∈ R4 the state-vector containing the internal state of the controller. A, B, C and D are small dense matrices with double precision coefficients (they define the behavior of the controller). The associated code is executed for 100 iterations. At each iteration, 7 elements of an array are computed. When the computation is completely carried out in half precision, the number of correct digits estimated by CADNA varies from 0 to 3 and its mean value is about 2.0. Table 2 presents the results provided by PROMISE with three precisions (half, single, double): it shows the number of requested digits, the number of executions performed, the number of variables of each precision in the code provided by PROMISE, the execution time of PROMISE. A mixed precision type configuration that includes half precision is found if 1 or 2 correct digits are requested in all the results. If 3 correct digits are requested, a single precision version of the code is provided. Then, the number of double precision variables increases, as the requested accuracy increases.

306

F. J´ez´equel et al. Table 2. PROMISE results for the signal processing application # req. digits # exec # half-# single-# double time (s)

6.3

1-2

58

6-12-0

58.16

3

52

0-18-0

51.47

4

55

0-15-3

47.53

5

62

0-11-7

50.92

6

67

0-9-9

53.76

7

66

0-7-11

50.89

8

63

0-4-14

47.36

9-11

52

0-1-17

38.10

Conjugate Gradient Code

In this experiment, the sequential version of a CG (Conjugate Gradient) code from the Seoul National University NPB (NAS Parallel Benchmarks) suite6 is analyzed using PROMISE. The code solves a linear system with a matrix of size 7,000 with 8 non-zero values per row by performing 15 CG iterations. Table 3 presents the results provided by PROMISE, the number of requested digits on the solution vector varying from 1 to 12. Indeed, because of rounding errors, the maximal accuracy is 12 correct digits. Whatever the requested accuracy from 1 to 12 correct digits, PROMISE always provides in less than 5 min a mixed-precision configuration that includes half precision variables. Out of the 325 (more than 847 billions) possible different configurations, PROMISE considers in the worst case only 100 possibilities. Table 3. PROMISE results for the Conjugate Gradient code # req. digits # exec # half-# single-# double time (s)

6

1

44

19-6-0

212.71

2

55

18-7-0

235.07

3

53

17-8-0

241.90

4

69

14-11-0

209.08

5

67

12-13-0

197.04

6-7

74

12-13-0

204.96

8

100

10-13-2

256.29

9

89

11-9-5

225.77

10

89

12-5-8

219.10

11

94

9-9-7

233.45

12

82

11-3-11

207.51

http://aces.snu.ac.kr/software/snu-npb.

Numerical Validation of Half Precision Simulations

7

307

Conclusion and Perspectives

User-friendly and efficient tools for the numerical validation of half precision codes and floating-point auto-tuning including half precision have been presented. They can ensure that a given code can be tuned to use low precision types while providing a user-defined numerical code quality. This work can be extended to other floating-point formats, such as bfloat16 available for instance on Google Cloud TPUs and ARM NEON CPUs. Since CADNA and PROMISE have been successfully used for the numerical validation of real-life applications in single or in double precision, we plan to control the numerical quality of large scale applications in half precision, such as deep learning codes. Acknowledgments. The authors gratefully acknowledge the financial support under the scope of the COMET program within the K2 Center “Integrated Computational Material, Process and Product Engineering (IC-MPPE)” (Project No 859480). This program is supported by the Austrian Federal Ministries for Transport, Innovation and Technology (BMVIT) and for Digital and Economic Affairs (BMDW), represented by the Austrian research funding association (FFG), and the federal states of Styria, Upper Austria and Tyrol.

References 1. Carson, E., Higham, N.J.: Accelerating the solution of linear systems by iterative refinement in three precisions. SIAM J. Sci. Comput. 40(2), 817–847 (2018) 2. Eberhart, P., Brajard, J., Fortin, P., J´ez´equel, F.: High performance numerical validation using stochastic arithmetic. Reliable Comput. 21, 35–52 (2015) 3. Graillat, S., J´ez´equel, F., Wang, S., Zhu, Y.: Stochastic arithmetic in multiprecision. Math. Comput. Sci. 5(4), 359–375 (2011) 4. Graillat, S., J´ez´equel, F., Picot, R., F´evotte, F., Lathuili`ere, B.: Auto-tuning for floating-point precision with discrete stochastic arithmetic. J. Comput. Sci. 36, 101017 (2019) 5. Haidar, A., et al.: The design of fast and energy-efficient linear solvers: on the potential of half-precision arithmetic and iterative refinement techniques, vol. 10860, pp. 586–600. Springer, Wuxi, China (2018). http://www.netlib.org/ utk/people/JackDongarra/papers.htm#2018. https://link.springer.com/chapter/ 10.1007%2F978-3-319-93698-7 45 6. IEEE Computer Society: IEEE Standard for Floating-Point Arithmetic. IEEE Standard 754-2008, August 2008 7. Lefebvre, D., Chevrel, P., Richard, S.: An H∞ based control design methodology dedicated to the active control of longitudinal oscillations. IEEE Trans. Control Syst. Technol. 11(6), 948–956 (2003) 8. Tagliavini, G., Marongiu, A., Benini, L.: Flexfloat: a software library for transprecision computing. IEEE Trans. Comput. Aided Des. Integr. Circuits Syst. 39(1), 145–156 (2020) 9. Vignes, J.: Discrete stochastic arithmetic for validating results of numerical software. Numer. Algorithms 37(1–4), 377–390 (2004) 10. Zeller, A.: Why Programs Fail, 2nd edn. Morgan Kaufmann, Boston (2009)

A Step Towards the Use of Chatbots to Support the Enterprise Decision-Making Processes Diogo Ferreira1 , Filipe Portela2,3(B) , and Manuel Filipe Santos2 2

1 Universidade do Minho, Guimar˜ aes, Portugal Algoritmi Research Centre, University of Minho, Braga, Portugal {cfp,mfs}@dsi.uminho.pt 3 IOTech - Innovation on Technology, Trofa, Portugal

Abstract. Chatbots are informatics applications that use Artificial Intelligence and natural language interfaces to simulate interactions with the end-user, providing a pleasant user experience. Generally, chatbots are related to customer service, trying either to answer the frequently asked questions that do not need direct human interaction or to generate leads to the after-client follow-up and contact. Chatbots represent a potential change in the way that people interact with information and online services. This article addresses the creation of a chatbot capable of generating insights for a manager on the decision-making process and the advantages that it represents. In order to create the referred chatbot, it is crucial to deal with multiple API’s interconnected, external services, and numerous databases from the organizations, always focused on the improvement of the business process like the time optimization and the quality of the information retrieved, and the user experience related to them. Further system analysis shows a time saving of approximately 60% compared to the traditional way to access the information. Keywords: Chatbot · Artificial Intelligence processing · API · Intranet

1

· Natural language

Introduction

The arising of new technologies has led to many application development changes and how they are used for management. One of the problems is related to the time managers/employees spend looking for important information. This information can be spread over several platforms and services that tend to increase, taking into account the increasing amount of services used by associations for the most diverse tasks. This constant need for the availability of information is seen as an example of a pervasive information system. Pervasive information systems refer to a paradigm where Information technology is embedded in the physical environment, capable of accommodating user needs and wants when desired [1]. Pervasive information systems differ from c The Author(s), under exclusive license to Springer Nature Switzerland AG 2021  ´ Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 308–317, 2021. A. https://doi.org/10.1007/978-3-030-72654-6_30

A Step Towards the Use of Chatbots in Decision-Making Processes

309

Information Systems (IS) based on the desktop paradigm. They encompass a complex, dynamic environment composed of multiple artifacts, capable of perceiving contextual information and supporting user mobility [2]. The chatbot created implements a pervasive information system capable of retrieving many relevant pieces of information to the end-user. According to Reshmi and Balakrishnan [3], chatbots are becoming increasingly popular in domains where human-computer interaction exists, such as virtual assistants, e-commerce, among others. So, if we can provide them with important tools to the business, saving crucial time in their day-to-day tasks. This article arose from the desire to find tools that support the management of an organization and, as far as possible, streamline tasks that facilitate the daily lives of managers so that they can focus themselves on tasks that move value and in which these cannot be replaced. While not entirely replacing the interaction with these tools through their interfaces, the use of chatbots can be of great importance in helping with small tasks that do not represent great complexity and in providing insights about an organization, reducing the ambiguity of the information obtained. Usually, this information from different tools is organized in dashboards, which can be problematic since its analysis, by itself, can generate some confusion or inadequate conclusions, especially when the number of business indicators and the number of tools used tend to increase over time. The use of chatbots to perform small tasks can be essential as it saves time in navigation between different applications. Thus, there is a need to create a scalable chatbot, in order to increase the speed with which users access information and work with it, promoting greater efficiency since they will not waste time with the crossing of data from a variety of sources. Consequently, and if the above points are found, it is expected that its use will transmit a pleasant and effective experience. This article also pretends to check if the development of an intelligent and functional chatbot that could interpret information from external services can improve and increase efficiency in multiple management or routine tasks. By explaining the benefits of the chatbots, it is expected to increase the receptivity to this type of technology, create an involvement with the users, and increase the company’s scalability requirements. The final goal is to motivate employees to adopt technologies, improve information and efficiency, and ensure uniformity and speed in service when applied to management positions. Besides this introduction, this document contains four more sections in its structure. The background sections contextualize business management and intranets, chatbots, and Artificial Intelligence in business and management, to demonstrate why creating a chatbot can be useful in an organizational context. The methods and tools section gives a general vision about the technologies used and why they are used, aiming to share knowledge about how the creation of a chatbot can be done. The results section shows some of the achieved results, how the system works as a whole, and the particularities that it presents. Finally, the conclusion section discusses the obtained results, how they can affect a company and its managers, and if it makes sense to apply this type of technology in business management.

310

2 2.1

D. Ferreira et al.

Background Business Management and Intranets

A company as an economic and social organization generates and handles a variety of information resources. The introduction of an intranet in the organization may facilitate the extensive sharing of resources and the efficient circulation of internal information, thus generating competitive advantage [4]. An intranet has a scope limited to the organization where it is applied and is built focused on the employees’ acceptance. The intranet uses a subset of service features that are usually used by the organization in order to make it simpler to use, operate, and manage [4], aiming mainly at increasing productivity and efficiency in tasks of the organization’s daily life. One of the fundamental pillars of an intranet is the simplification of the creation and dissemination of information. In addition to allowing the search and obtaining of information, it must always take into account its security, the control of reading and writing access, to keep the information reliable and integrated. These factors positively influence the mechanisms involved in exchanging information, guaranteeing broader, faster access and a higher quality of information, which strongly assists the decision-making process [5]. After analyzing the benefits and potential mentioned above, it is possible to realize the added value that an intranet can add to an organization. 2.2

Chatbots

Chatbots can be characterized as “systems online of human-computer dialogue with natural language” [6] to simulate a typical conversation between humans, which can be written, spoken, or both. Ideally, chatbots should be able to simulate a conversation in such a natural way that they can prevent the user from realizing that he is talking to a robot [7]. With the World Wide Web’s growth, it has become increasingly difficult for ordinary users to access quickly and effectively the information they need [8]. From this premise, the need arises to increasingly facilitate the interaction of the standard user with each system, which is the primary objective of using chatbots in organizations. In a simplified way, a chatbot has the function of receiving the user’s request, interpreting, analyzing and contextualizing the request to determine the reaction he will have to the request, and finally translating it into a relevant response [7]. According to Brandtzaeg and Følstad [9], Chatbots represent a potential change in the way people interact with online data and services. The interest of large companies in this technology is already well-known, as they are increasingly seen as an excellent means of attracting users through customer service and marketing text messages, overcoming the need for specific applications or web pages for that purpose. However, the transition to using chatbots as a means of interacting with the user is not easy. The design of new interactive technology requires much knowledge of the users’ motivation to use it, acting as a facilitator.

A Step Towards the Use of Chatbots in Decision-Making Processes

311

Market research was also carried out to perceive existing solutions. In this research, chatbots found had something in common with the result intended to be obtained in this project. However, none of the solutions found added across the board the characteristics intended for the final solution. The chatbot obtained in this project can guarantee authentication from the internal mechanisms of the company, to work with event triggers, to be integrated with different modules, API’s, such as event scheduling services, billing platforms, project management platforms, and databases used by the company and, also, to generate data analysis and control access to them according to the user level in question. These are the characteristics that were considered most important for the prototype developed. 2.3

Artificial Intelligence in Business and Management

Artificial Intelligence (AI) increasingly dominates business environments and even external environments. However, there are several criticisms due to the fear that AI will endanger people’s role in the future of business management and operations. However, it is tough to refute that AI has multiple positive impacts on most business operations and management. Therefore, investing in AI will help increase sustainability, and market leadership will be improved [10]. According to Zahrani and Marghalani [10], companies and people need to look at AI as support for humans rather than a replacement. AI systems can analyze vast amounts of data much more quickly than humans and create a form of action on their own through the generated insights, which can be used to improve organizations’ decision-making. According to Kolbjørnsrud et al. [11], Artificial Intelligence will become cheaper, efficient, and potentially more impartial than humans in the tasks mentioned above, which should not be a concern for managers. For Kolbjørnsrud et al. [11], managers should be able to be prepared and prepare organizations to work alongside AI systems. Still, there is a long way until people can trust this type of technology as a valuable ally without always putting in causes its operation and validity.

3

Methods and Tools

The chatbot development had some particularities in order to meet the specificities of this technology. From the beginning of this project, all the integrations with external platforms were made in the company Intranet’s API. It allowed the company to have all the tools, information, and integrations in one place, rather than splitting it into multiple places. The need for a standardization rase from the necessity to have all the integrations working under the same logic, allowing the reduction of maintenance effort and reuse of the integrations, reducing future projects’ development times.

312

D. Ferreira et al.

In the solution creation first phase, all external platforms’ integrations were made, always applying the modularity principle. This way, after the finish of all integrations, all the focus was on developing the intelligent chatbot. This development of the chatbot had two different approaches. The first approach was based on a solution that uses both Python programming language and the TensorFlow library. However, at this stage, it became clear that it was not a viable solution due to multiple reasons like the high processing and storage capacity required, the long processing time of the model training, and the lack of a training dataset in the company language. Given these reasons, this approach to creating the chatbot was discarded. Besides, subsequent research has revealed an promising solution, DialogFlow. At the outset, this platform would be able to comply with the distinction as a necessity of this project, which determines the decision to change the approach. Dialogflow [12] is a natural language processing platform that facilitates the design and integration of a conversational interface with applications for mobile devices, web applications, devices, bots, interactive voice response systems, among others [12]. This platform allows the creation of chatbots that may be adapted to the needs of companies. Google [12] highlights the following main characteristics: the Google machine learning technology capable of recognizing the user intention and extract predefined entities, such as time, date, and numbers; it facilitates the development of serverless applications; makes possible the use of company knowledge in the agent; is capable of automated telephone answering; multiplatform and multilingual; developed for use with voice control; capability to improve experiences with integrated analysis; ability to analyze the user’s feelings. The DialogFlow API was integrated into the Intranet API in the created solution, turning this one into its central focus. For this reason, it can have total control of access to different levels of permissions in obtaining information, thus ensuring increased protection in the use of chatbot in an organizational context. It is also important to note that this integration with the DialogFlow API works through sessions. When a request is made to the Intranet API, three parameters are mandatory: the token of access to the Intranet, the user’s message, and a unique universal identifier (UUID) for each session. This session identifier must always be the same, as long as the user has the Intranet application open, and must be renewed only when reaccessed. Thus, it is possible to understand whether the user’s conversation with the chatbot remains the same or not, allowing the definition of the context it is taking.

4

Results

The main objective has been achieved with this project: obtaining a functional chatbot that allows the execution of operations that assist the management of companies. However, it is essential to view the whole panorama with a critical perspective to fully understand the final solution. Following is the flow of the operation of the compliance check operation to give a better sense of the process of the same events.

A Step Towards the Use of Chatbots in Decision-Making Processes

313

Fig. 1. Intranet API integration flow. Adapted from Google DialogFlow [13]

Figure 1 represents the workflow of the integration with DialogFlow. The figure’s steps are described below, relating them to an operation to check for nearby events to better understand the entire flow’s functioning. In the first phase, the end-user types an expression, such as “Do I have anything scheduled for tomorrow?”. The service, in this case, a chatbot interface, sends that expression from the end-user to the Intranet API. In turn, the intranet verifies the user’s authenticity and forwards the intention detection request message to the Dialogflow, that is, the previous example message. Dialogflow detects the corresponding intent and sends a response message to the Intranet API. This message contains information about the related intent, the action, the parameters, and the response defined for that same intention. For the example provided, the intent would be corresponding to the event verification operation. The parameter would be data for the next day, since DialogFlow can extract dates from words like “tomorrow”. The defined answer can be an example of a final response or, if the search parameter was not provided and it is defined as mandatory, a sentence that would question the end-user about the data to which he was referring, such as “In which day what positions do you want?”. The Intranet API verifies that the user who made the request has access to the type of intent identified and, if so, takes the necessary measures, such as database queries or external API calls, to obtain the information required to satisfy that intent. The example above would contact the Google Calendar API to get the events of the requested day. The Intranet API sends a response to the end-user, which, in the example shown, corresponds to the events existing in Google Calendar. In this way, the Intranet API has the central focus of the system. Therefore, it can have total control of access to different levels of permissions in obtaining information, thus ensuring increased protection when using the chatbot in an organizational context.

314

D. Ferreira et al.

Next, a comparison is shown between the average time it takes the user to perform some operations on the proprietary applications and the built chatbot. In this way, it is possible to perceive the improvement in terms of time that this technology represents for its users. Table 1. Comparison of average time spent in some operations. Operation

Normal time Chatbot time App

Get upcoming events

10 s

Get invoices issued

30 s

8s

Moloni

73.30%

Searching files by name

20 s3

8s

Google Drive

60.00%

12 s

Google Sheets

52.00%

Get employees presences 25 s

5s

% fast

Google Calendar 50.00%

5s FactorialHR Record of daily activity 15 s 3 It may vary on the number of files and their organization.

66.60%

The times shown do not consider the authentication on the respective platforms. If we consider the authentication process, the times shown should increase by 20 s or more on average to the table’s “Normal Times” values. Given the example of the “Get upcoming events” operation, the normal time wasted is 10 s, which turns into approximately 30 s if we consider the authentication process. On the other hand, the chatbot time is only 5 s, traducing in an improvement of 50%, which may vary due to the user typing speed and intent detection at the first interaction. As seen from the analysis of Table 1, the introduction of a chatbot can significantly reduce the time required to perform various operations. The average percentage of improvement is 60% faster. Although the figures presented may seem relatively low, after some operations, significant time in saving the use of platforms is being saved. Below in Table 2, is shown some examples of interaction with the built chatbot. Table 2. Example of interaction with the chatbot. User’s question example

Chatbot answer

Record my entry into the company, please

Entry registered at 9:00 am

What was the last invoice from customer X?

The last invoice from customer X was issued on (date) with the value of (value) euros

Do I have any pendent tasks?

For the project (project name) you have to do the task (task name) until (due date)

From the analysis of Table 2, it is possible to understand the type of interactions available between the end-user and the chatbot and the ease of use for routine tasks.

A Step Towards the Use of Chatbots in Decision-Making Processes

5

315

Discussion

The main objective has been achieved with this project: obtaining a functional chatbot that allows the execution of operations that assist companies’ management. The SWOT analysis in Fig. 2 shows the main factors considered in this reflection.

Fig. 2. SWOT analysis.

The SWOT analysis presented in Fig. 2 gives an account of the main strengths, weaknesses, opportunities, and threats that the developed system presents. It is presented as strengths the fact that this is a differentiating and innovative solution, as there are no reports of the use of chatbots in the fulfillment of functions to assist business management; be a modular and reusable solution, as it is possible to reuse both the integrations made with external services and the chatbot itself, changing only parameters related to access credentials; be able to perform multiple operations on the various external services that have been integrated; be easy to use and interact with the chatbot and the features it presents and which was, since early, a concern in the development of this project; be agile in obtaining information, because an employee of the organization is able to obtain information related to different services in one place. This information can eventually be acquired by crossing data from multiple platforms, which manually would represent a long waiting time. In the case of weaknesses, some appear as the limitation of interaction, since through the text interactions become challenging to perform more complex operations and require a large amount of information. Another weakness presented is the need to add training phrases to make the chatbot more accurate in detecting intents and, consequently, in the answers provided. Although the

316

D. Ferreira et al.

system has been trained to deal with one hundred and one intents, this does not equal human dialogue capacity. Therefore, the limited capacity for dialogue was pointed out as a weakness. The opportunities pointed out are: the possibility for the whole system to be developed to a new level in terms of the features provided and the accuracy of the chatbot itself; the possibility of chatbot becoming an important tool in the management of the company, facilitating the routine tasks of employees. Finally, the threats found consisted of the lack of use of this solution, which may lead to inaccurate results since the chatbot is not frequently trained. These inaccurate results may lead to the abandonment of the chatbot’s use since if the few times that users try to take advantage of this tool are unsuccessful, they will reach a certain point that they will no longer try to use it. The main objective of obtaining the functional prototype was successfully achieved. Although there are features to be optimized, which are only possible to detect with exhaustive use, it was possible to obtain the functional chatbot that performs many operations and interacting with external platforms.

6

Conclusion

With the main objective completed, there is a need to study how chatbot’s introduction can help business management. The first advantage found right from the start is related to the time saved in using this tool. Instead of a user accessing different platforms and authenticating in all of them to perform small operations, with this solution, it is possible to have these same platforms centralized and perform these same operations via chatbot. As shown earlier in Table 1, the time saved in some operations is up to 60% on average (not considering the authentication process), which traduces to a significant amount of time saved. Another significant advantage is the quality of the information obtained and the possibility of improving it over time. As the chatbot operation has inherent integration with external services, it is possible to get data on each of them in real-time. Many platforms are integrated on the Intranet, allowing multiple aggregations and crossings of data, leading to increasingly powerful insights about the organization. When made available through the chatbot, these insights allow obtaining information and indicators in real-time of the entire company’s functioning. However, a different type of work, which could be studied in the context of the use of chatbot by the organization’s employees, would be the implementation of the concept of gamification in the interaction with this tool. The implementation of a points distribution system would have the ability to reward employees who perform specific tasks according to certain configurable standards. Through this score, the company’s employees could be seconded and awarded after a certain time period. Thus, it would be possible to guarantee that the chatbot would have regular use, making it more robust and precise in its operation. The present article is an introductory article that defines the initial development strategy. As future work, and to deepen this project’s development, an

A Step Towards the Use of Chatbots in Decision-Making Processes

317

extended version article was created called “The Enterprise Decision-Making Processes using Chatbots - A case study”. In this article, details such as the functional requirements, the solution architecture in terms of adopted components, and the chatbot’s implementation and interaction are discussed in depth. Acknowledgments. This work has been supported by FCT – Funda¸ca ˜ para a Ciˆencia e Tecnologia within the R&D Units Project Scope: UIDB/00319/2020. This project was also supported by IOTech - Innovation on Technology trough an internal scholarship.

References 1. Fernandes, G., Portela, F., Santos, M.F.: PWA and Pervasive Information System – A New Era, vol. 1161 AISC, pp. 334–343. Springer, Cham (2020) 2. Kourouthanassis, P., Giaglis, G., Karaiskos, D.: Delineating ‘pervasiveness’ in pervasive information systems: a taxonomical framework and design implications. J. Inf. Technol. 25, 201 (2010) 3. Reshmi, K.S., Balakrishnan: Empowering chatbots with business intelligence by big data integration. Int. J. Adv. Res. Comput. Sci., 9(1), 627–631 (2018) 4. Pinto, F.Q., de Carvalho, P.M.: As intranets como valor acrescentado para uma empresa. I Encontro Nacional de Engenharia e Gest˜ ao Industrial, pp. 1–6 (1997) 5. Pedriali, M.C., Cuccurullo, D., Bol¸cone, V.K.: O uso da Intranet como ferramenta para a dissemina¸ca ˜o da gest˜ ao do conhecimento nas organiza¸co ˜es. IX Congresso Internacional de Custos (2005) 6. Cahn, J.: CHATBOT: Architecture, Design, & Development. Ph.D. thesis, University of Pennsylvania - School of Engineering and Applied Science (2017) 7. Smartsheet, “Ultimate Guide to Artificial Intelligence Chatbots — Smartsheet,” (2019) 8. Deshpande, A., Shahane, A., Gadre, D., Deshpande, M., Joshi, P.M.: A survey of various chatbot implementation techniques. Int. J. Comput. Eng. Appl. XI, ISSN 2321-3469, p. 7 (2017) 9. Brandtzaeg, P.B., Følstad, A.: Why people use chatbots. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10673 LNCS, November, pp. 377–392 (2017) 10. Zahrani, A., Marghalani, A.: How Artificial Intelligent Transform Business?. Saudi Aramco, Information Technology (2018) 11. Kolbjørnsrud, V., Amico, R., Thomas, R.J.: Leadership development: how artificial intelligence will redefine management. Harvard Business Review, pp. 1–6 (2016) 12. “Dialogflow — Google Cloud,”. https://cloud.google.com/dialogflow/docs. Accessed 2 Oct 2020 13. “Documenta¸ca ˜o do Dialogflow,”. https://cloud.google.com/dialogflow/es/docs/ api-overview. Accessed 2 Oct 2020

COVID-19’s Impact on the Telecommunications Companies Latifah Almuqren1,2(B) and Alexandra I. Cristea2 1 Princess Nourah Bint Abdul Rahman University, Riyadh, Saudi Arabia

[email protected] 2 Durham University, Durham, UK

Abstract. Now the world is witnessing most significant challenges due the Covid19 crisis. Beyond health effects, it has social and economic effects. With the enormous amount of data available and the widespread use of social web globally, research can and should use it to provide solutions. Customer satisfaction is known to affect customer churn (customers leaving companies), which is a problem affecting many companies in competitive and volatile markets – like the current one. One easily available open source of customer opinions are tweets – more relevant now in the online world. Whilst Natural Language Processing (NLP) on tweets is not new, few studies target customer satisfaction, and NLP body of research on Arabic tweets is modest; we are not aware of any other study on this during a global pandemic. Our research thus aims to propose a new model based on Twitter mining to measure customer satisfaction during Covid-19, as well as compare customer satisfaction before and during the crisis. This is a use case for the largest Telecom companies in Saudi Arabia, and we involve the popular method of Sentiment Analysis (SA) for the task. We additionally propose a new Saudi lexicon and apply it to monitor real-time customer satisfaction on Twitter using three different transfer network models on Arabic sentiment analysis. Also, this research evaluates using these models on Arabic Sentiment Analysis as the first study comparing between three different transfer network models on Arabic text. Keywords: Sentiment analysis · Customer Satisfaction · Arabic

1 Introduction With the growing use of social media sites worldwide over the last ten years, SA has recently become a prominent and useful technique for capturing public opinion in many different disciplines. Sentiment analysis or ‘opinion mining’ refers to the computational processing of opinions, feelings and attitudes towards a particular event or issue [21]. SA has a vital function for real-life applications and decision-making process in various domains [3]. In the context of the global Covid-19 pandemic and the quarantine, there were several impacts on different sector, one of the critical sectors is Telecommunication (telecom) [6]. As an inevitable consequence of Covid-19 a lot of people used the telecom service for © The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 Á. Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 318–327, 2021. https://doi.org/10.1007/978-3-030-72654-6_31

COVID-19’s Impact on the Telecommunications Companies

319

working from their home or for entertainment. Which constituted a burden on the telecom companies to increased network flexibility and this affected the telecom investments [6]. Several reports issued to addressing the impact of Covid-19 on telecom companies [6, 19]. On other hand, there is a lack of studies addressed the customer satisfaction toward telecom company services during the Covid-19 and the quarantine. Therefore, our study aims to find out what impact Covid-19 had on the customer satisfaction toward telecom company services and the implications for potential decisions. While, the sentiment have a crucial role in many decisions, including business-related ones. The detection of sentiment polarity, however, is a challenging task, due to the sentiment resources limitations in different languages. It remains a largely unexplored research area for the Arabic language [3]. This is due to the complexity of Arabic [3]. It has many forms, which are Classical Arabic (CA), as in the book of Islam’s Holy Quran, Modern Standard Arabic (MSA) used in newspapers, education and formal speaking, Dialectical Arabic (DA) which is informal everyday spoken language, found in chat rooms, and social media platforms. Additionally, DA forms differs from one Arab country to another. Thus, unfortunately, resources used for one Arabic country cannot be applied to another. Hence, DA analysis, targeted here, is complicated, requiring a native speaker. This paper attempts to alleviate this matter by focusing on Arabic Sentiment Analysis and provide solutions to one of the challenges that faces Arabic SA by creating a Saudi dialectical data set extracted from Twitter. In addition, we evaluate applying the transformer network a Robustly Optimized BERT Pretraining (RoBERTa) [16] on the Arabic data set and compare it to AraBERT [5] and hULMonA [10] as two transfer models designed for Arabic language. We chose RoBERTa because it was shown to outperform Bidirectional Encoder Representations (BERT) in sentiment classification tasks [16], although BERT is the best in many NLP tasks [16]. To the best of our knowledge, this is the first attempt of using RoBERTa for Arabic sentiment analysis. Moreover, this paper predicts customer satisfaction before and during Covid-19, in order understand how the pandemic and its traceable effects has influenced customers as well as telecoms companies. The main contributions of this paper have measured the impact of a pandemic (Covid-19) and quarantine on customer satisfaction toward telecom company services, compare customer satisfaction before and during the crisis and compare for the first-time different transfer network models RoBERTa, AraBERT, and hULMonA on Arabic Sentiment Analysis.

2 Related Research a. Customer Satisfaction Customer satisfaction (CS) is defined as the outcome of using a service, resulting from the comparisons that the buyer makes with other similar providers in terms of the rewards and costs of the service [11]. Social media includes a variety of platforms that allow people to share and exchange information, producing an abundance of data. Literature [13, 23] showed though that mining social media data is important for both marketers and customers, as—1) it provides a wealth of information about customers for the company [13], 2) it enables the extraction of information hidden in data that can benefit businesses

320

L. Almuqren and A. I. Cristea

by exploring their customers’ views about particular products or services, 3) it helps to develop a recommendation system for maintaining existing customers or gaining new ones, and 4) it is also useful for building confidence among customers and stakeholders [23]. However, only a few studies measure customer satisfaction, via social media mining and all these studies mined English tweets [18]. Mostafa [18] mined random tweets on Twitter to find the consumer’s sentiments towards certain brands through sentiment analysis. The findings proved that there is a positive consumer sentiment towards famous brands. Both studies used the sentiment analysis to measure the satisfaction. In addition, both studies measured the customer satisfaction in business field using English tweets. b. Using Transformer Language Model with Sentiment Analysis Recently, pre-trained language models have achieved good results with different NLP target tasks, due to ability of these models to learn with few parameters [10]. Previous approaches depended on features [12]. The Universal Language Model Fine-tuning (ULMFiT) pre-trained language model [12], composed of three ASGD Weight-Dropped Long Short-Term Model (AWD-LSTM) and an LSTM layer [17], is perfectly accurate on different NLP tasks. The newest language model is BERT [9]. It used a Transformer network [28]. BERT outperformed the other pre-trained language models, due to its ability to manipulate context from both directions. Another pre-trained language model is RoBERTa [16], which is an enhanced version of the BERT model [9]. It trained on larger data, with longer series, and is the best model on the General Language Understanding Evaluation (GLUE) benchmark [24], Stanford Question Answering Dataset (SQuAD) [20] and ReAding Comprehension from Examinations (RACE) [14]. c. Using Transformers with Arabic Sentiment Analysis (ASA) The use of transfer language models is still new for ASA studies. There is a lack of studies has been published so far [2, 5, 10]. Al-Twairesh and Al-Negheimish [2] used the BERT model [9] on an Arabic tweets dataset. Their generic and sentiment-specific word-embedding model outperformed the BERT model. They explained that this was because the BERT model was trained on Wikipedia, which is written in MSA, whereas dialects are used on Twitter. in Addition, there is a first Arabic universal Language model hULMonA [10], it based on ULMFiT. The results showed that hULMonA achieved stateof-art in Arabic sentiment analysis. The most recent Arabic universal Language model is AraBert [5].

3 Data Processing a. Data Sets To build the data sets, we used Python to fetch Arabic tweets based on certain search keys. The hashtags used in the search were the ones that indicated different Saudi telecom companies: STC, Mobily and Zain. Then we grabbed the top hashtags mentioning these

COVID-19’s Impact on the Telecommunications Companies

321

telecom companies, which were: #STC, #Mobily, #Zain, and . The aim was to monitor the telecom customers’ sentiments continuously. The raw data set comprised 3.5 million Arabic tweets. Then, we chose the sample of Saudi tweets randomly from the data set to construct our first corpus, AraCust1, Table 1. Our AraCust1 corpus is comprised of 20,000 Saudi tweets collected randomly from the data set, representing the customer satisfaction situation before Covid-19. The second corpus, which we called AraCust2, has been collected during Covid-19 from second of March until second of May, the period from registering the first case of Covid19 in Saudi Arabia, Table 1. This corpus represents the customer satisfaction situation during Covid-19. Thus, we filtered tweets based on user location and time-zone, to identify Saudi tweets. Table 1. Companies and the total number of unique tweets for AraCust1and AraCust2 Company

Twitter handle

# of unique tweets AraCust1

AraCust2

STC

@STC_KSA, @STCcare

7,590

7000

Mobily

@Mobily, @Mobily1100

6,460

4004

Zain

@ZainKSA, @ZainHelpSA

5,950

6000

20,000

17,004

Total

To clean the data set, we eliminate the non-Arabic tweets. Also, re-tweets were discarded. We filter all the features that unnecessary and will decrease the classifiers accuracy e.g., links, user mention, punctuation marks, and stop words. Then, applying pre-processing producers on the data set (tokenisation and normalisation). Normalization such as removing Kishida and uniting the same letters with different shapes. The cleaning and pre-processing were done using Toolkit (NLTK) library in Python. Examples before & after pre-processing (AraCust1) are shown in Table 2. Table 2. Subset of the AraCust1 corpus before pre-processing Tweet in Arabic before pre-processing

Label

Company

Negative

STC

Positive

Mobily

Tweet in English Change the Company Thank you

After pre processing

322

L. Almuqren and A. I. Cristea

b. Annotation for Corpora We need to add a new aspect to the data set, which is the sentiment label. We used the binary labels (Positive, Negative) for both corpuses. Each label expresses the strongest emotion for each tweet, following previous recommendations [3]. This means that a given tweet might have other emotions associated, which we discard for the time being. The annotation process was done manually for the AraCust1 corpus by three annotators that have the experience with the annotation process. Every annotator needs to assign one label per tweet for the whole data set. To identify the difficulty of the annotation task, the inter-annotator agreement (IAA) was used, to measure the trustworthiness of the annotation scheme. If there were more than two annotators, then Fleiss’ Kappa [8] is recommended to measure the reliability of the annotation process. Therefore, we applied it, obtaining a value of 0.50, based on the level of acceptance, which is considered moderate [15]. Kappa, k, can be defined as: P − Pe (1) 1 − Pe The factor 1-Pe computes the degree of agreement that is attainable above chance, and P-Pe represents the degree of agreement achieved above chance. If the annotators are in complete agreement, then k = 1. If there is no agreement, then k < 0. The AraCust2 corpus annotation done using AraSTw lexicon that created and evaluated by the authors. k=

4 Model Construction a. Evaluation Metrics To evaluate the performances of the models, we used four metrics suitable for binary classification [7]: micro averages of (Precision (Pr), Recall (Rc), and F1), Accuracy (Ac). The micro average is suitable for binary-classes, especially if the classes are imbalanced, because the micro-average totals the contribution of all classes to the average metric calculation [22]. It aggregates the precision and recall of the classes. b. RoBERTa Model Construction We used RoBERTa model, BERT based, using the parameters seed = 42 to build on the random weight, the precision floating for GPU, the batch size = 16 and 64 for the maximum sequence length. After visualizing the number of tokens in most tweets, the maximum tokens per tweet is 30 tokens. Therefore, we padded all tokens up to this size. After that, the model converts the word to an integer. The model using discriminative fine-tuning and gradual unfreezing. The model freezes all the groups but the two last ones. It used the minimum numerical gradient: 1.00E-03 and minimum loss divided by 10: 3.02E-04. To implement the sentiment analysis task, it used discriminative fine-tuning and gradual unfreezing. That means to predict the next token, based on the present series of tokens in the sentiment corpus, with various learning rates, from 1e-02 to le-06. After that, the model unfreezes the output layer, after each epoch, layer by layer. It freezes all the groups except for the last two layers.

COVID-19’s Impact on the Telecommunications Companies

323

c. AraBERT Model Construction It is BERT based model; it trained on different Arabic datasets. It used the BERT basic configuration [9]. Except, it added a special pretraining prior the experiment specific to Arabic Language. It used “‫“ “ال‬Al” before the word, it is a prefix without meaning, by using a Fast and Accurate Arabic Segmenter (Farasa) [1] to segment the word. d. HULMonA Model Construction HULMonA is the first Universal Language Model (ULM) for Arabic language. It is based on ULMFiT [12]. It pretrained on Large Arabic corpus and fine-tuned to many tasks. It consists of three stages: 1. training AWD-LSTM model [17] on Arabic Wikipedia corpus, 2. finetuning the model on a destination corpus, 3. and for text classification, they included a classification layer on the model.

5 Experiment Results, Discussion and Evaluation a. Model Results Comparing the results of using RoBERTa, AraBERT, and hULMona models using the micro average of different metrics, Table 3 shows that the results favour to the AraBERT model with 94.0% accuracy. Table 3. Comparing between RoBERTa, AraBERT, and hULMonA models Model

Accuracy F1

Recall Precision

RoBERTa model

92.1

92.2 92.0

91.1

AraBERT model

94.0

92.6 92.1

93.0

hULMonA model 90.8

79.8 89.0

84.0

b. Discussion To get the reasons behind the obtained results, we analysed the three models’ architecture. AraBERT outperformed the two others models due to: 1. It trained on different Arabic data sets Modern Standard Arabic data sets and dialectical data sets. 2. It applied a special pretraining specific to Arabic Language. 3. It used Farasa [1] a pre-processing tool directed to the Arabic language; It outperformed the state-of-art MADAMIRA [2]. Although, hULMonA trained on different Arabic data sets, it performed worse than other models, because it based on ULMFiT [12] which is one directional model. Additionally, it lacks to appropriate pre-processing for Arabic text.

324

L. Almuqren and A. I. Cristea

Regarding RoBERTa which a BERT based model, it transfers each Arabic letter to Latin character and every Latin character face one integer number. That is the reason for decreasing the performance of the model. c. Predicting Customer Satisfaction We used AraBERT model which achieved the higher result to predict the customer satisfaction for the AraCust1 corpora based on the predefined companies STC, Mobily and Zain. First, we calculated the customer satisfaction as follows: cust_sat = total_ratings /(2*num_customers). where: num_customers = len(ratings) total_ratings = sum(ratings) (the summation of all ratings) rating: binary rating. Then we divided the corpus based on the company. We calculated the average accuracy of predicted customer satisfaction using the model. Our result showed that the predicted customer satisfaction (using the model) for the three companies STC, Mobily and Zain before Covid-19, 33.27%, 29.88% and 333.24% were below 50%. d. Evaluate using the Sentiment Analysis Approach This study has used a sentiment analysis to design a new real-time model to measure customer satisfaction. To evaluate the sentiment analysis approach is effective to measure customer satisfaction, it has used a questionnaire to the same customers that we mined their tweets to evaluate the approach by comparing the predicted customer satisfaction (using the AraBERT model) with actual customer satisfaction (using the survey). It can be seen in Table 4 that the predicted and actual customer satisfaction rates are approximate. That results denote to efficiency of using sentiment analysis approach to achieve the goal of predicting the customer satisfaction of telecom companies based on the Twitter analysis. These results should encourage the decision-makers to consider using Twitter analyses for measuring customer satisfaction and to include it as a new method for evaluating their marketing strategies. Table 4. Rate of predicted customer satisfaction vs actual customer satisfaction Company

Predicted customer satisfaction

Actual customer satisfaction

STC

33.27%

20.1%

Mobily

29.88%

22.89%

Zain

33.24%,

22.91%

COVID-19’s Impact on the Telecommunications Companies

325

6 Predicting Customer Satisfaction During Covid-19 After proving the efficiency of using AraBERT model on the Arabic dialect textspecially- Saudi tweet corpus in term of applying sentiment analysis approach to predict the customer satisfaction; we used lexicon-approach to annotate the AraCust2. We used the generated AraSTw lexicon to annotate AraCust2 using binary sentiment (positive, negative). Secondly, we inspected AraCust2 manually, to check the tweets’ sentiment; the positive tweets were more than the negative tweets. After that, we applied AraBERT to predict the customer satisfaction as it performed best with the AraCust1 corpus. The accuracy was 90% and Favg = 88.8, as in Table 5. Table 5. Evaluation results on AraCust2 using AraBERT model. Label Precision Recall F1-score F avg

Accuracy

0

90.0

82.0

85.5

90.0

1

89.0

95.0

92.0

88.8

7 Comparing Between Customer Satisfaction Percentage Before and During Covid-19 Using the AraBERT Model Because of the pandemic and its effects such as the quarantine, it enforced people to work from home and to connect with the outsider world depending on the technology. As a result, the telecom industry became an essential sector. Which constituted a burden on the telecom companies to offer a reliable and fast data and voice services. The Covid19 pandemic made people to use the telecom services more than before. This resulted to increase the benefits of the Saudi telecom companies to 7% in the first quarter of 2020 [4]. As we noticed there are several reports that studied the impact of Covid-19 on the telecom sector from investment side [6, 19]. This is not the case in our study; it focused on the customers side toward the telecom services during Covid-19 by measuring the customer satisfaction toward the telecom companies in Saudi Arabia. Our result showed that comparing the predicted customer satisfaction (using the AraBERT model) before and during Covid-19 (Table 6), the customer satisfaction percentage for the three companies STC, Mobily and Zain before Covid-19, 33.27%, 29.88% and 33.24%, were below 50%. However, the customer satisfaction during Covid-19 and quarantine was raised to 58.99%, 50.05% and 62.50% for STC, Mobily and Zain, respectively, Fig. 1.

326

L. Almuqren and A. I. Cristea

80.00% 60.00% 40.00% 20.00% 0.00% STC

Mobily

Zain

Customer’s Satisfaction Before Covid-19 Customer’s Satisfaction After Covid-19 Fig. 1. Comparing between Customer satisfaction before and during Covid-19

Table 6. Percentage of customer’s satisfaction before and during Covid-19 in Saudi Arabia. Company Customer’s satisfaction before Covid-19 Customer’s satisfaction during Covid-19 STC

33.27%

58.99%

Mobily

29.88%

50.05%

Zain

33.24%

62.50%

8 Conclusion This study proposes a new model based on Twitter mining to measure customer satisfaction during Covid-19. To find out if Covid-19 has an impact on the telecom companies. Interestingly, we found that customer satisfaction toward telecom companies enhanced during Covid-19, because of home quarantine and growing demand for Internet and voice services. Additionally, we used the RoBERTa model for the first time for Arabic sentiment analysis and compared it to AraBERT and hULMonA models for Arabic Sentiment Analysis. Datasets were collected before and during Covid-19. Surprisingly, results showed that AraBERT outperformed RoBERTa, and hULMonA based on several measures. A reason behind the obtained results may be due to the RoBERTa didn’t train on Arabic dataset and hULMonA depending on forward layer rather than bidirectional layers. Still, initial results for hULMonA are promising, and further experimentations are planned to use. Acknowledgment. This research was funded by the Deanship of Scientific Research at Princess Nourah bint Abdulrahman University, through the Pioneer Research Funding Program (Grant No# PR-1441-3).

COVID-19’s Impact on the Telecommunications Companies

327

References 1. Abdelali, A., et al.: Farasa: a fast and furious segmenter for arabic. In: Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations (2016) 2. Al-Twairesh, N., Al-Negheimish, H.: Surface and Deep Features Ensemble for Sentiment Analysis of Arabic Tweets (2019) 3. Al-twairesh, N.S.: Sentiment Analysis of Twitter: A Study on the Saudi Community (2016) 4. Awalan. Saudi Arabia: Corona pushes profits of telecom companies up (2020). https:// www. Accessed 1 June 2020 awa lan. com/ Art icle/ 5130/ 5. Baly, F., Hajj, H.: AraBERT: transformer-based model for Arabic language understanding. In: Proceedings of the 4th Workshop on Open-Source Arabic Corpora and Processing Tools, with a Shared Task on Offensive Language Detection (2020) 6. Casey, M., Wigginton, C.: Understanding the sector impact of Covid-19 Telecommunications. Deloitte (2020) 7. Chicco, D., Jurman, G.: The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation. BMC Genomics 21(1), 6 (2020) 8. Davies, M., Fleiss, J.L.: Measuring agreement for multinomial data. Biometrics 38, 1047–1051 (1982) 9. Devlin, J., et al.: Bert: pre-training of deep bidirectional transformers for language understanding (2018) 10. ElJundi, O., et al.: hULMonA: the universal language model in Arabic. In: Proceedings of the Fourth Arabic Natural Language Processing Workshop (2019) 11. Gustafsson, A., Johnson, M.D., Roos, I.: The effects of customer satisfaction, relationship commitment dimensions, and triggers on customer retention. J. Mark. 69, 210–218 (2005) 12. Howard, J., Ruder, S.: Universal language model fine-tuning for text classification (2018) 13. Kennedy, H.: Perspectives on sentiment analysis. J. Broadcast. Electron. Media 56(4), 435–450 (2012) 14. Lai, G., et al.: Race: large-scale reading comprehension dataset from examinations (2017) 15. Landis, J.R., Koch, G.G.: Landis_Jr__Koch_Gg_1977_Kappa_and_Observer_Agreement. Biometrics 33, 159–174 (1977) 16. Liu, Y., et al.: Roberta: a robustly optimized bert pretraining approach (2019) 17. Merity, S., Keskar, N.S., Socher, R.: Regularizing and optimizing LSTM language models (2017) 18. Mostafa, M.M.: More than words: social networks’ text mining for consumer brand sentiments. Expert Syst. Appl. 40(10), 4241–4251 (2013) 19. Veligura, N., et al.: COVID-19’s Impact on the Global Telecommunications Industry. International Finance Corporation (IFC), May 2020 20. Rajpurkar, P., et al.: Squad: 100,000+ questions for machine comprehension of text (2016) 21. Ravi, K., Ravi, V.: A survey on opinion mining and sentiment analysis: tasks, approaches and applications. Knowl.-Based Syst. 89, 14–46 (2015) 22. Sokolova, M., Lapalme, G.: A systematic analysis of performance measures for classification tasks. Inf. Process. Manag. 45(4), 427–437 (2009) 23. Tundjungsari, V.: Business Intelligence with Social Media and Data Mining to Support Customer Satisfaction in Telecommunication Industry, vol. 1 (2013) 24. Wang, A., et al.: GLUE: a multi-task benchmark and analysis platform for natural language understanding (2018)

Link Prediction Using Double Degree Equation with Mutual and Popular Nodes Haji Gul1(B) , Adnan Amin1 , Furqan Nasir2 , Sher Jeel Ahmad2 , and Muhammad Wasim2 1

2

Center for Excellence in Information Technology, Institute of Management Sciences, Peshawar, Pakistan [email protected], [email protected] City University of Science and Information Technology, Peshawar, Pakistan [email protected]

Abstract. Link prediction among the objects of data is one of the essential tasks in several application fields, such as spam mail detection, expert detection, information retrieval, influence detection, recommender system, surveillance of communication, and disease prediction. Link prediction aims to estimate the missing link based on the existing links in the given complex network. To solve the link prediction problem, different techniques have been introduced. The similarity-based link prediction techniques are effective, but still imperfect. To better the predicting accuracy, we introduce a link prediction technique that aims to improve the accuracy of the existing link prediction problem by combining the mutual nodes or neighbor with the feature or information of popular nodes. We also examine the proposed technique empirically and lead extensive tests over complex network datasets obtained from different domains. The experimental results show that the proposed technique has performed well as compared to other state-of-the-art techniques. The average result of the proposed link predictor is 0.8812 while others are CN 0.8734, Katz 0.8112, LHN 0.7778, PA 0.7367, and PD accuracy is 0.6538. Keywords: Link prediction · Mutual nodes · Popular nodes Complex networks · Prediction and recommendation

1

·

Introduction

For several years, real-world complex network structure and shape have been analyzed, where the majority data of real-world complex network are still deficient. Social network relationships between people may be simple and intentionally unobserved. The interaction of species or cells can be estimated through a high-priced experiment in real-world networks such as Twitter, Facebook [1]. It empowers social entertainer to accomplish different jobs, i.e., switching views, sharing information, and also can memorize from different other social c The Author(s), under exclusive license to Springer Nature Switzerland AG 2021  ´ Rocha et al. (Eds.): WorldCIST 2021, AISC 1368, pp. 328–337, 2021. A. https://doi.org/10.1007/978-3-030-72654-6_32

Link Prediction Using Double Degree Equation with Mutual

329

networks entertainers [2]. People’s interaction at public places like university pupils, restaurant clients, and cafe customers, are examples of real-world networks. These also can be expressed by online social networks i.e., Twitter and Facebook [3]. In the field of graph theory, online real-world social networks can be represented by social graph. People and the interaction/relationship between peoples are express by nodes and edges respectively in a social graph. There are many indices, that are used to predict link based on the node, topological [4], social theory, and many other hybrid link predictors [5]. Link prediction contains different techniques and applications in many complex networks [6]. From the past decade, online social networks have become a vital segment of our daily lives, and give opportunities to interact and stay combine. The link prediction problem rotate into a model to compare real-world complex network design [7]. Therefore, it provides many opportunities for researchers to examine the different views of peoples/nodes’ behaviors as well as the attributes of social graphs or networks. Analyzing social graph is an unusual work and facing lots of difficulties. Many problems concerning social graphs or social networks are studied, such as detection of community, examining the structure of a real-world complex network, and visualization of the network. While link prediction is also an important problem of the social graph/network, i.e., to analyze a missing link or future link in the time duration from t to t+1 utilizing the idea of network excavation/mining. Suppose a social complex network given in Fig. 1, there are total five nodes and edges at time t, while at time t+1 Fig. 2 a new link predicted and expressed by red dotted line. Link prediction applications exist in many fields like recommendation citation, recommender system, co-authorship network, analyzing social network of criminal, and recommender system of items. Related forecast system present for understudies intellectual abilities estimation [8].

2

Overview of Link Prediction

An extra addressable issue is to perceive the edge between a pair of nodes. How new associations are made? Particularly, the problem imparted here is the expected likelihood of a link between two vertices. Later on, this is named as the link prediction problem [9]. Assume an undirected system G = (V, E) where the set of hubs/vertices/ nodes addressed by V and set of connections/edges/links by E. If an association exists among x and y express by e = (x, y) at that point the response worth will be 1 regardless of 0 in the notice infectiousness system. Edge prediction plans to distinguish the edges between nodes dependent on past information. A system may be undirected or directed. At this moment will consider the undirected system, where loop and numerous edges are ignored. Total possible links are a piece of very common undirected network information. All possible links in an undirected system can be signified by G=

n(n − 1) 2

(1)

330

H. Gul et al.

In the formula n = V address the amounts of vertices or nodes in set V. Assume a given complex system in Fig. 1 at time t, we need to investigate the future connection at time t + 1. Let us think about the co-authorship system of researchers, we can anticipate all conceivable edges that will shape later on. In this given system, there are a total of four vertices/nodes. Assume V 2 needs to create a connection with another vertex in the given system, here vertex V 2 is the source node and the predicted node will be the destination node. In Fig. 2, V 2 has been anticipated an edge with V 4 based on the common neighbor. V 2 likewise can make a connection with V 3 yet dependent on common neighbor there is only one common neighbor, following CN procedure that must not be the predicted link.

Fig. 1. A Complex Network

2.1

Fig. 2. Link Predicted

Significance of the Problem

Real-world complex network analysis has become an important computer science research problem. It is quietly known that analyzing real-world social networks is complicated because these networks have a large number of nodes with millions of users or maybe more, while links are billions in numbers. These all real-world systems of networks are also recognized by complex networks [10] and their mechanisms are examined. All the mechanisms provide a lot of benefits for many disciplines, like detection of community, protein-protein interaction in real-world sociology bio-informatics networks, recommendation system and discover collaboration between scientists or conceivable friends suggestions. Consequently, the link prediction problem has become a significant work in real-world network analysis. 2.2

Topological Based Link Prediction

Note that, node-based indices need many features to operate, while topological based indices work without any parameter of vertices and edges. Topological based methods use the topology information. In recent decades multiple topological based indices are introduced. Stat-of-the-art link prediction indices are enumerated below.

Link Prediction Using Double Degree Equation with Mutual

331

Common Neighbour (CN) [11]: The higher sum of neighbors the maximum possibility that there will be a link between x and y. CN (x, y) = |Γ(x) ∩ Γ(y)|

(2)

In the given formula T(x) and T(y) express the numbers of neighbors of nodes x and y. Jaccard Coefficient (JC) [12]: JC (Jaccard Coefficient) normalizes the size of the common neighbor. The pair of nodes with a higher proportion will have a larger weight. |Γ(x) ∩ Γ(y)| JC(x, y) = (3) |Γ(x) ∪ Γ(y)| Adamic Adar (AA) [13]: The AA describes that, node pairs who have lesser neighbors have higher weight. The formula is given below: 

AA(x, y) =

w∈{Γ(x)∩Γ(y)}

1 , log |Γ(w)|

(4)

in the given equation w express common neighbour of x and y. Preferential Attachment (PA) [14]: Preferential Attachment suggests that recent edges will more likely join vertices having a larger degree. P A(x, y) = |Γ(x)| · |Γ(y)|

(5)

Katz Index[15]: The Katz link prediction method deliberate based on 3 rout lengths. It captures all paths among vertices into supposition. Katz’s higher the length reduces the weight of resultant similarities. Katz(u, v) =

∞ 

  β l · pathl uv

(6)

l=1

pathlxy express the set of a whole path from source node x to the destination node y with the distance of l,0