Entertainment for Education. Digital Techniques and Systems: 5th International Conference on E-learning and Games, Edutainment 2010, Changchun, China, ... (Lecture Notes in Computer Science, 6249) 3642145329, 9783642145322

Citation preview

Lecture Notes in Computer Science Commenced Publication in 1973 Founding and Former Series Editors: Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen

Editorial Board David Hutchison Lancaster University, UK Takeo Kanade Carnegie Mellon University, Pittsburgh, PA, USA Josef Kittler University of Surrey, Guildford, UK Jon M. Kleinberg Cornell University, Ithaca, NY, USA Alfred Kobsa University of California, Irvine, CA, USA Friedemann Mattern ETH Zurich, Switzerland John C. Mitchell Stanford University, CA, USA Moni Naor Weizmann Institute of Science, Rehovot, Israel Oscar Nierstrasz University of Bern, Switzerland C. Pandu Rangan Indian Institute of Technology, Madras, India Bernhard Steffen TU Dortmund University, Germany Madhu Sudan Microsoft Research, Cambridge, MA, USA Demetri Terzopoulos University of California, Los Angeles, CA, USA Doug Tygar University of California, Berkeley, CA, USA Gerhard Weikum Max-Planck Institute of Computer Science, Saarbruecken, Germany

6249

Xiaopeng Zhang Shaochun Zhong Zhigeng Pan Kevin Wong Ruwei Yun (Eds.)

Entertainment for Education Digital Techniques and Systems 5th International Conference on E-learning and Games, Edutainment 2010 Changchun, China, August 16-18, 2010 Proceedings

13

Volume Editors Xiaopeng Zhang Chinese Academy of Sciences, Beijing, China E-mail: [email protected] Shaochun Zhong Northeast Normal University, ChangChun, China E-mail: [email protected] Zhigeng Pan Zhejiang University Hangzhou, China E-mail: [email protected] Kevin Wong Murdoch University, South St. Murdoch, WA, Australia E-mail: [email protected] Ruwei Yun Nanjing Normal University, Nanjing, China E-mail: [email protected]

Library of Congress Control Number: 2010931021 CR Subject Classification (1998): K.3.1, H.5.2, I.2.6, H.4, I.3.7, H.5.1 LNCS Sublibrary: SL 3 – Information Systems and Application, incl. Internet/Web and HCI ISSN ISBN-10 ISBN-13

0302-9743 3-642-14532-9 Springer Berlin Heidelberg New York 978-3-642-14532-2 Springer Berlin Heidelberg New York

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting, reproduction on microfilms or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. springer.com © Springer-Verlag Berlin Heidelberg 2010 Printed in Germany Typesetting: Camera-ready by author, data conversion by Scientific Publishing Services, Chennai, India Printed on acid-free paper 06/3180

Preface

With the technical advancement of digital media and the medium of communication in recent years, there is a widespread interest in digital entertainment. An emerging technical research area edutainment, or educational entertainment, has been accepted as education using digital entertainment. Edutainment has been recognized as an effective way of learning using modern digital media tools, like computers, games, mobile phones, televisions, or other virtual reality applications, which emphasizes the use of entertainment with application to the education domain. The Edutainment conference series was established in 2006 and subsequently organized as a special event for researchers working in this new interest area of e-learning and digital entertainment. The main purpose of Edutainment conferences is to facilitate the discussion, presentation, and information exchange of the scientific and technological development in the new community. The Edutainment conference series becomes a valuable opportunity for researchers, engineers, and graduate students to communicate at these international annual events. The conference series includes plenary invited talks, workshops, tutorials, paper presentation tracks, and panel discussions. The Edutainment conference series was initiated in Hangzhou, China in 2006. Following the success of the first event, the second (Edutainment 2007 in Hong Kong, China), third (Edutainment 2008 in Nanjing, China), and fourth editions (Edutainment 2009 in Banff, Canada) were organized. Edutainment 2010 was held during August 16–18, 2010 in Changchun, China. Two workshops were jointly organized together with Edutainment 2010. The two workshops focused on topics in Digital Resources for Innovative Teaching and Learning Methods, and in the Theory and Practice of E-learning and Game-Based Learning Environments. For Edutainment 2010, we received 222 submissions from 27 different countries and regions, including Australia, Austria, Belgium, Brazil, Canada, China, Denmark, France, Germany, Greece, Hong Kong (China), Hungary, India, Italy, Japan, Republic of Korea, Malaysia, Mexico, The Netherlands, Portugal, Singapore, Spain, Sweden, Taiwan (China), Tunisia, UK, and USA. A total of 63 papers were selected after peer reviews for this volume. Each paper submitted was reviewed at least by two reviewers in Edutainment 2010. The topics of these papers fall into six different areas ranging from fundamental theories and techniques, tools and systems development, and applications. These topics include E-Learning Tools and Platforms, E-Learning System for Education, E-Learning Environments and Applications, Game Techniques for Edutainment, Multimedia Techniques for Edutainment, and Computer Animation and Graphics for Edutainment. This book constitutes the refereed proceedings of the 5th International Conference on E-learning and Games, Edutainment 2010, held in Changchun, China, in August 2010.

VI

Preface

We are grateful to the International Program Committee and the reviewers for their great effort and commitment to quality in getting all the papers reviewed in a short period of time. We are grateful to the Organizing Committee for their tireless efforts supporting this event. We would also like to thank the authors and participants for their enthusiasm and contribution to this conference.

May 2010

Xiaopeng Zhang Shaochun Zhong Zhigeng Pan Kevin Wong Ruwei Yun

Organization

Organizing Committee Conference Chairs BodoUrban Zhigeng Pan Thanos Vasilakos

FhG - IGD, Rostock, Germany Zhejiang University, China University of Western Macedonia, Greece

Program Chairs Xiaopeng Zhang Stephane Natkin Kevin Wong

Institute of Automation, CAS, China CNAM/ENJMIN, France Murdoch University, Australia

Workshop Co-chairs Dongdai Zhou Xiaochun Cheng Yongjiang Zhong

Northeast Normal University, China Middlesex University, UK Northeast Normal University, China

Financial Chair Shuzhen Song

Northeast Normal University, China

Publication Chair Ruwei Yun

Nanjing Normal University, China

Publicity Chair Junjie Shang Abdennour El Rhalibi

Peking University, China Liverpool John Moores University, UK

Organizing Chairs Shaochun Zhong

Northeast Normal University, China

VIII

Organization

General Secretary Zhuo Zhang

Northeast Normal University, China

International Program Committee Sangchul Chul Ahn Isabel Machado Alexandre Daniel Aranda Dominique Archambault Ruth Aylett Rafael Bidarra Paul Brna Judy Brown Eliya Buyukkaya Yiyu Cai Christophe Chaillou Tak-Wai Chan Maiga Chang Yam San Chee Tian Chen Xiaochun Cheng Jinshi Cui Akshay Darbari Weiming Dong Stephane Donikian Abdennour El Rhalibi Zhiyi Fang Guangzheng Fei Marco Furini Stefan Göbel Martin Goebel Ong Sing Goh Mohamed Hamada Gaoqi He Michitaka Hirose KinChuen Hui Wu-Yuin Hwang Marc Jaeger

Korea Institute of Science and Technology, South Korea University Campus of Lisbon, Portugal Universitat Oberta de Catalunya, Spain University Pierre et Marie Curie, France Heriot-Watt University Edinburgh, UK Delft University of Technology, The Netherlands University of Glasgow, UK Brown Cunningham Associates, USA University of Pierre and Marie Curie (Paris 6), France Nanyang Technological University, Singapore, Singapore ALCOVE, INRIA Futurs, France National Central University, Taiwan, China Athabasca University, Canada National Institute of Education, Singapore Shanghai Dianji University, China Middlesex University, UK Peking University, China Digital Imaging & New Media, India LIAMA-NLPR, Institute of Automation, CAS, China Bunraku research team/INRIA Rennes, France Liverpool John Moores University, UK Jilin University, China Communication University of China, China University of Piemonte Orientale, Italy Technische Universität Darmstadt, Germany FleXilution, Germany Murdoch University, Western Australia, Australia University of Aizu, Japan East China University of Science and Technology, China University of Tokyo, Japan Chinese University of Hong Kong, Hong Kong, China National Central University, Taiwan, China INRIA, France

Organization

Xiaogang Jin Bin Shyan Jong Bill Kapralos Börje Karlsson Mike Katchabaw Gerard Jounghyun Kim Sehwan Kim Kinshuk Frank Klassner Chen-Wo Kuo Mona Laroussi Fong-Lok Lee Jong Weon Lee Elinda Lee James Lin Craig Lindley Qingtang Liu Zhen Liu Bruce Maxim Chunhui Mei Wolfgang Mueller Ryohei Nakatsu Yoshihiro Okada Zhigeng Pan Ming-Yong Pang Daniel Pletinckx Edmond Prakash Jon Preston Manjunath Ramachandra Matthias Rauterberg Theresa-Marie Rhyne Marco Roccetti Timothy Roden Demetrios Sampson Cecilia Sik Lanyi Mohd Shahrizal Sunar Bing Tang Ruck Thawonmas Harold Thwaites

IX

Zhejiang University, China Chung-Yuan Christian University, Taiwan, China University of Ontario Institute of Technology, Canada Pontifical Catholic University of Rio de Janeiro (PUC-Rio), Brazil University of Western Ontario, Canada Pohang University of Science and Technology (POSTECH), South Korea University of California, Santa Barbara, USA Athabasca University, Canada Villanova University, USA National Palace Museum, Taiwan, China INSAT/USTL, Tunisia The Chinese University of Hong Kong, Hong Kong, China Sejong University, South Korea Murdoch University, Western Australia, Australia National Palace Museum, Taiwan, China University of Gotland, Sweden Huazhong Normal University, China Beijing Normal University, China University of Michigan-Dearborn, USA SAMSUNG Research, USA University of Education Weingarten, Germany National University of Singapore, Singapore Kyushu University, Japan Zhejiang University, China Nanjing Normal University, China Visual Dimension, Belgium Manchester Metropolitan University, UK Southern Polytechnic State University, USA Philips Electronics, India Eindhoven University of Technology, The Netherlands North Carolina State University, USA University of Bologna, Italy Angelo State University, USA University of Piraeus, Greece University of Pannonia, Hungary Universiti Teknologi Malaysia, Malaysia US Imaging, CGGVeritas, Houston, Texas, USA Ritsumeikan University, Japan Multimedia University, Cyberjaya, Malaysia

X

Organization

Feng Tian Clark Verbrugge Charlie Wang Dai-Yi Wang Yangsheng Wang Yinghui Wang Wenyong Wang Toyohide Watanabe Kevin Wong Woontack Woo Xihong Wu Yueguang Xie Gang Yang Ruwei Yun Xiaopeng Zhang Zhuo Zhang Shaochun Zhong Yongjiang Zhong Dongdai Zhou Jiejie Zhu

Bournemouth University, UK McGill University, Canada Chinese University of Hong Kong, Hong Kong, China Providence University, Taiwan, China Institute of Automation, CAS, China Xi'an University of Science and Technology, China Northeast Normal University, China Nagoya University, Japan Murdoch University, Australia GIST U-VR Lab., Korea, South Peking University, China Northeast Normal University, China Beijing Forestry University, China Nanjing Normal University, China Institute of Automation, CAS, China Northeast Normal University, China Northeast Normal University, China Northeast Normal University, China Northeast Normal University, China University of Central Florida, USA

Sponsoring Institutions The success of The 5th International Conference on E-learning and Game was due to the financial and practical support from various institutions. Sponsor: z VR Committee, China Society of Image and Graphics Co-sponsors: z National Science Foundation of China (NSFC) z International Journal of Virtual Reality (IJVR) z Journal of Information and Computing Science (JICS) z Transactions on Edutainment (ToE) z Northeast Normal University, China z Zhejiang University, China z Nanjing Normal University, China z LIAMA-NLPR, Institute of Automation, CAS, China We would like to show our appreciation to the above sponsor and co-sponsors for offering the opportunity and support to organize Edutainment 2010 with a diversified scientific and social program.

Table of Contents

E-Learning Tools and Platforms Effect of Multimedia Annotation System on Improving English Writing and Speaking Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Wu-Yuin Hwang, Rustam Shadiev, and Szu-Min Huang

1

Smap: To Generate the Personalized Learning Paths for Different Learning Style Learners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Juan Yang, Hongtao Liu, and Zhixing Huang

13

Computer-Supported Collaborative Conceptual Change . . . . . . . . . . . . . . . Xiaodong Xu and Yingjie Ren

23

Optimization Technique for Commercial Mobile MMORPG . . . . . . . . . . . Jianmin Wang, Zibin Zheng, Peter Tam, and Jianping Liu

34

Design and Implementation of TCP/IP Protocol Learning Tool . . . . . . . . Feng Li and Nana Yu

46

A Framework for Creating, Training, and Testing Self-Organizing Maps for Recognizing Learning Styles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ram´ on Zatarain-Cabada, M.L. Barr´ on-Estrada, Viridiana Ponce Angulo, Ad´ an Jos´e Garc´ıa, and Carlos A. Reyes Garc´ıa Simulating Dynamic Evolvement of Collective Learning Behaviors Based on Voronoi Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Xiang-min Gao and Ming-yong Pang SPICEreading: A Three-in-One Share Platform in Cooperative English Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yu-Ju Lan, Yao-Ting Sung, Sheng-Kuang Chiu, Chia-huei Lin, Hsien-Sheng Hsiao, Tzu-Chien Liu, and Kuo-En Chang The Design and Implementation of Middle School Physics Optical Simulation Experiment Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Xiangchun Ma, Shaochun Zhong, Da Xu, and Chunhong Zhang Research on the Establishment of Structural E-Learning Resources . . . . . Zhuo Zhang, Wei Wang, Zhongwu Zhou, and Yongbin Chen Research on Virtual Experiment Intelligent Tutoring System Based on Multi-agent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Xianye Li, Fahui Ma, Shijun Zhong, Lin Tang, and Zhongwei Han

53

65

74

84 92

100

XII

Table of Contents

A Model-Driven Architecture Approach for Developing E-Learning Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Xiao Cong, Hongmei Zhang, Dongdai Zhou, Peng Lu, and Ling Qin

111

E-Learning System for Education Knowledge Preference Based Learning Community Construction and Service Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hua Deng, Yongzhao Zhan, and Qirong Mao

123

Developing an Online History Educational System to Present the Progression of Spatial Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jia-Jiunn Lo and Hsiao-Han Tu

135

A Bibliometric Study of E-Learning Literature on SSCI Database . . . . . . Johannes K. Chiang, Chen-Wo Kuo, and Yu-Hsiang Yang Pedagogical Strategy Model in Adaptive Learning System Focusing on Learning Styles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hongxia Liu, Wei Zhao, and Ming Liang

145

156

Transferring Design Knowledge: Challenges and Opportunities . . . . . . . . . Jun Hu, Wei Chen, Christoph Bartneck, and Matthias Rauterberg

165

The Content Balancing Method for Item Selection in CAT . . . . . . . . . . . . Peng Lu, Dongdai Zhou, Xiao Cong, Wei Wang, and Da Xu

173

The Formative Evaluation’s Impact on Online Learning . . . . . . . . . . . . . . . Mei Pu and Lu Wang

185

Psychological Perspectives on Social Behaviors of Chinese MMORPG Players . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ge Qian Research on the Adaptive Strategy of Adaptive Learning System . . . . . . . Lian Bian and Yueguang Xie Research on an Educational Software Defect Prediction Model Based on SVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Guang-jie Liu and Wen-yong Wang Webgame Based Collaborative Learning Design: A Case Study . . . . . . . . . Jie Jian, Yueguang Xie, Wenhe Tang, and Chunhui Wang

192 203

215 223

E-Learning Environments and Applications Design of a Medical Simulator Hard- and Software Architecture . . . . . . . . P. Peters, F. Delbressine, and L. Feijs

235

Table of Contents

Design and Implementation of Semantic Matching Based Automatic Scoring System for C Programming Language . . . . . . . . . . . . . . . . . . . . . . . Jinrong Li, Wei Pan, Ren Zhang, Feiquan Chen, Shenglong Nie, and Xiaoming He

XIII

247

An Analysis Framework of Activity Context in e-Learning Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yanlin Zheng, Luyi Li, and Fanglin Zheng

258

Distributed Cognition and Ecological Field of Learning in Network Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kan Tao

269

A Multimodal Virtual Anatomy E-Learning Tool for Medical Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jianfeng Lu, Li Li, and Goh Poh Sun

278

To Construct the Architecture of Digital Learning Port for Free Normal Students and Analyze the Impact on Teacher Education . . . . . . . . . . . . . . Yi Zhang, Chao Du, Ge Dong, and Fan Zhang

288

Node Localization for Distributed Simulation Based on Logical Node Group in Simulation Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hai Huang, Lei Tian, Wei Wu, Songlin Sun, and Xiaojun Jing

298

Using Graph Edit Distance to Diagnose Student’s Science Process Skill in Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ming-Xiang Fan, Maiga Chang, Rita Kuo, and Jia-Sheng Heh

307

Intelligent Assessment in Math Education for Complete Induction Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Wolfgang M¨ uller and Maren Hiob-Viertler

317

Research on the Method of Recomposing Learning Objects and Tools in Adaptive Learning Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pan Xie, Longmei Ye, Yueming Huang, Youwei Chen, and Liwu Lin

326

A Study of Formative Assessment Index System for Educational Technology Competence Based on AHP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kefei Wang and Lu Ming

337

Research of Automatic Assessment System of Virtual Experiment in Middle School Biology Based on the Virtual Simulation Technology . . . . Yuxi Wang, Shaochun Zhong, Haoran Zhang, Yongjiang Zhong, and Ling Bai Resource Organization and Management of the Platform for Supporting Teacher Education Innovation Based on IPv6 . . . . . . . . . . . . . . . . . . . . . . . . Dongxue Liu, Zhen Liu, Lin Liu, and Yun Ren

345

353

XIV

Table of Contents

Game Techniques for Edutainment A Glissade on the Learning Curve: Multi-adaptive Immersive Educational Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Michael D. Kickmeier-Rust, Christina M. Steiner, Elke Mattheiss, and Dietrich Albert

361

Experimental Development of Competitive Digital Educational Games on Multi-touch Screen for Young Children . . . . . . . . . . . . . . . . . . . . . . . . . . . Xiaohua Yu, Mian Zhang, Jie Ren, Huifang Zhao, and Zhiting Zhu

367

Strategy Research about Exploiting the Attention Resource of Learners in Educational Games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Li Qian, Sujing Zhang, and Ke Jin

376

Planning Serious Games: Adapting Approaches for Development . . . . . . . Alana M. Morais, Herbet F. Rodrigues, Liliane S. Machado, and Ana Maria G. Valen¸ca

385

UML Modeling for Software System of Edu-Game . . . . . . . . . . . . . . . . . . . . Yufang Sun and Ruwei Yun

395

A Common Software Architecture for Educational Games . . . . . . . . . . . . . Wenfeng Hu

405

O3D-Based Game Learning Environments for Cultural Heritage Online Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lu Wang, Jian-wei Guo, Cheng-lei Yang, Hai-seng Zhao, and Xiang-xu Meng Simulator and Robot-Based Game for Learning Automata Theory . . . . . Mohamed Hamada and Sayota Sato Personalized, Adaptive Digital Educational Games Using Narrative Game-Based Learning Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stefan G¨ obel, Viktor Wendel, Christopher Ritter, and Ralf Steinmetz

417

429

438

Multimedia Techniques for Edutainment Virtual Classrooms Supporting a Two-Way Synchronized Video and Audio Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Xin Li, Minghua Li, and Liren Zeng

446

Optimal Bi-directional Seam Carving for Content-Aware Image Resizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Meiling Shi, Guoqin Peng, Lei Yang, and Dan Xu

456

Real-Time Hand Gesture Recognition Based on Vision . . . . . . . . . . . . . . . . Yu Ren and Chengcheng Gu

468

Table of Contents

A Vertical Search Engine Based on Visual and Textual Features . . . . . . . Kun Wu, Hai Jin, Ran Zheng, and Qin Zhang Hand Gesture Recognition in Natural State Based on Rotation Invariance and OpenCV Realization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Baoyun Zhang, Ruwei Yun, and Huaqing Qiu Robust Hand Posture Recognition Integrating Multi-cue Hand Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chuanbo Weng, Yang Li, Mingmin Zhang, Kangde Guo, Xing Tang, and Zhigeng Pan Spectrally-Based Single Image Relighting . . . . . . . . . . . . . . . . . . . . . . . . . . . Xiaoxiong Xing, Weiming Dong, Xiaopeng Zhang, and Jean-Claude Paul Multiple Layer Displacement Mapping with Lossless Image Compression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Youngjae Chun, Sunyong Park, and Kyoungsu Oh

XV

476

486

497

509

518

Computer Animation and Graphics for Edutainment Research on Shadow Map Based Shadow Generation . . . . . . . . . . . . . . . . . Jie Guo, Xiao-Yang Xu, Yan Zhuang, and Jin-Gui Pan A Case for Web-Based Interactive 3D Game Using Motion Capture Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Su Cai, Qiang Liu, and Luyi Li Sketch-Based Instancing of Parameterized 3D Models . . . . . . . . . . . . . . . . . Dan Xiao, Zhigeng Pan, and Renzhong Zhou Digital Animation: Repercussions of New Media on Traditional Animation Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Filipe Costa Luz

529

541 550

562

Towards Virtual Actors for Acting Out Stories . . . . . . . . . . . . . . . . . . . . . . . Ido A. Iurgel, Rog´erio E. da Silva, and Manuel F. dos Santos

570

Progressive 3D Model Compression Based on Surfacelet . . . . . . . . . . . . . . . Jinjiang Li and Hui Fan

582

An Improved Artificial Potential Field Algorithm for Virtual Human Path Planning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Junwen Sheng, Gaoqi He, Weibin Guo, and Jianhua Li

592

Research on Collision Detection Algorithm Based on Particle Swarm Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Wei Zhao, Li-Jun Li, and Cheng-Shou Chen

602

XVI

Table of Contents

Parallel Collision Detection Algorithm Based on OBB Tree and MapReduce . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Wei Zhao, Cheng-Shou Chen, and Li-Jun Li

610

Creation of Tree Models from Freehand Sketches by Building 3D Skeleton Point Cloud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jia Liu, Xiaopeng Zhang, Hongjun Li, and Mingrui Dai

621

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

633

Effect of Multimedia Annotation System on Improving English Writing and Speaking Performance Wu-Yuin Hwang, Rustam Shadiev, and Szu-Min Huang Graduate Institute of Network Learning Technology at the National Central University, No.300, Jhongda Rd., Jhongli City, Taoyuan County 32001, Taiwan (R.O.C.) {wyhwang,rshadiev,smhwang}@cl.ncu.edu.tw

Abstract. In this study learning activities were designed with the focus on students’ English writing and speaking performance. The Virtual Pen (VPen), multimedia web annotation system were provided to help student participate at learning activities by creating annotations, sharing them and giving feedback to peers’ work. One experiment was conducted with VPen in an English class for a period of one semester and the following results were obtained. Students perceived VPen as easy to use, useful during participation at learning activities and students had positive attitude toward using VPen. Besides, students found designed activities as useful in improving their writing and speaking performance and playful. Furthermore, students’ actual VPen usage had significant correlation with speaking and writing performance. Further investigation demonstrated that students’ speaking and writing performance significantly correlated with learning achievement. Based on our finding we conclude that learning activities designed in this study with VPen system employed facilitate students’ writing and speaking skill and therefore improve their learning achievement. Keywords: Computer Assisted Language Learning, Virtual Pen, Playfulness, Language Productive Skills.

1 Introduction Since English currently became the most wide spread language, learning it as a second language become a major trend in many non-English-speaking country, including Taiwan. In English learning, four skills such as listening, speaking, reading and writing are divided into two functions, language input (receptive skills) and language output (productive skills); the former includes listening and reading, the last includes writing and speaking. It is suggested to keep balance between those two functions in order to promote language learning efficiently. Lin [21], Liu and Littlewood [22] and other scholars already reported that students from East Asia, including Taiwan get to use to such a learning method which focus on language input and they have less opportunities for language output functions. As a result students feel unease and anxiety during active learning, especially at group discussions, asking questions or giving feedback to others in class. X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 1–12, 2010. © Springer-Verlag Berlin Heidelberg 2010

2

W.-Y. Hwang, R. Shadiev, and S.-M. Huang

In this study, the online English learning activities were designed with emphasis on improving students’ productive skills, i.e. writing and speaking. Students have participated at learning activities by creating or completing the stories related to subject; they have added annotations with textual or audio content. Pictures related to subject and students’ everyday life were provided for student to increase their imagination and interests to English learning. Designed Virtual Pen (VPen), multimedia web annotation system was provided to students to help them participating at learning activities, like create annotations, share them with the peers and give a feedback to peers’ work. In this study we tried to investigate students’ perception toward designed learning activities and VPen, students’ actual usage of the system and its influence on students’ writing and speaking performance. Besides we analyzed the relationship between students’ writing and speaking performance and its influence on learning achievement. The paper is organized as follows. First, literature survey on English learning skills and interactive language learning are reviewed. Then the theories of Computer Assisted Language Learning and Technology Acceptance Model are discussed to guide the analysis. The discussion of the underlying method in the study was followed. Then, the results and analysis of the study are shown. Finally, discussions and conclusions are given.

2 Literature Review Harmer [12] suggested that the in-class language learning activities can be divided into two functions: the language input when information is stored in students’ brains and the language output when students apply information that they learned. According to Krashen [19] listening and reading belong to the receptive skills and speaking and writing classified into the productive skills [26]. Harmer [12] proposed keeping balance between those two functions in order to promote language learning efficiently. In any learning process, communication and interaction are considered as important elements [9]. The researches in language learning are emphasis the relationship between students’ active responses during learning process and their acquired proficiency [11]. In interactive language learning student receives the contents of learning material from tutor, peer or any other resource, response to the question or complete assignment and get feedback; this cycle of interaction promote students’ language skills [3], [29]. Ellis [10] argued that if students actively participate in learning process, i.e. response to questions, complete assignment and receive or provide corrective feedback their learning achievements can be positively affected. Make students participate at language learning process actively is a challenge that the teachers face [2], especially in East Asia. Song [24], Liu and Littlewood [22] in research of East Asian students learning English found students’ low confidence in their ability to speak, initial unease with group discussions and the extreme anxiety generated simply by the thought of asking a question in class. Huang and Lee [13] argued that students in the traditional Taiwanese language learning classroom didn’t get to use to interact with other students, share individuals’ ideas and support their viewpoints.

Effect of Multimedia Annotation System on Improving English Writing

3

Computer Assisted Language Learning (CALL) is a tool that helps to facilitate language learning process. According to Warschauer [27] the current phase of the development of CALL is an integrative approach and it based on multimedia computers and the Internet. One way that computers and internet are used in language learning is to support communication, synchronous and/or asynchronous. Huang [13] and Lamy [20] with their colleagues proposed introducing asynchronous discussion into language learning classroom to promote students’ interaction and active participation, like share ideas and support viewpoints. They argued that the asynchronicity of online interactions allow participants to have flexible time to reflect on topics before they comment or carry out online tasks. Kiesler [18] and Sproull [25] with their colleagues suggested that asynchronous discussion can reduce students’ worry and anxiety while learning target language, then makes the student learn actively. According to Warschauer [27] multimedia technology allows a variety of media, like text, graphics, sound etc. to be accessed. Provision of multimedia in language learning stimulates students’ imagination to give meaningful output. Caldwell [4] argued the asynchronous environment that provides threaded discussion, combined with creative use of Internet resources and multimedia, forms the most effective combination for successful student engagement; it increases students’ interaction and motivation, particularly for language learning. Web annotation systems were proposed to better utilize digital content. Students could maximize learning from multimedia learning materials on the web by using web annotation tools. Hwang and his colleagues [14], [15], [16] designed multimedia web annotation system VPen and implemented it in series of experiments. The results showed that the system was both useful and effective and students that used VPen system performed significantly better than the students that didn’t use the system. Davis [6] proposed Technology Acceptance Model (TAM) that evaluates the effect of system characteristics on user acceptance of computer based information system. Later, in 1992 Davis and his colleagues [7] were studied users’ motivation to use technology. They defined extrinsic motivation as the motivation to perform an activity because it is perceived to produce valued outcomes that are distinct from the activity itself and intrinsic motivation which referred to the performance of an activity for no apparent reinforcement other than the process of performing the activity per se. An example of intrinsic motivation is Perceived Playfulness [17] towards the system. Barnett [1] argued that individuals with playful dispositions are said to be guided by internal motivation, an orientation toward process with self-imposed goals, a tendency to attribute their own meanings to objects or behaviors, a focus on pretense and non literality, a freedom from externally imposed rules, and active involvement. Moon & Kim [23], Csikszentmihalyi [5] and Deci [8] identified four constructs of “playfulness” (a) the degree of student’s attention, (b) the degree of student’s curiosity induced by the activity, (c) the degree of students’ enjoyment and pleasure while participating in the activity, and (d) the degree of students’ feeling of joyful while learning in a company.

4

W.-Y. Hwang, R. Shadiev, and S.-M. Huang

In this study we designed learning activities that stimulate students’ productive skills, facilitate their writing and speaking and encourage students’ interaction. Besides, VPen multimedia web annotation system was introduced to help English learning, especially in improving writing and speaking skills. Students used VPen to create annotations with textual and audio content, share annotations with peers and interact by asking or answering questions and giving a feedback. TAM theory was applied to explore students’ attitude toward the learning activities and the system.

3 Research Design 3.1 The Subject and the Participants of the Research Twenty seven third-grade students from one junior high school participated in this study. The study was carried out during the first semester from September until December, 2006. During the experiment students visited computer classroom every two weeks to participate at learning activities and complete their assignments. 3.2 VPen Multimedia Annotation System The main interface of VPen multimedia web annotation system is shown in Figure 1. Students had to login into web site first using individual user name and password. We uploaded English learning materials on VPen and the system provides students with a structured method to navigate through English learning activities, as shown on the left side of Figure 1. In order to move to specific sections of content, students click corresponding links. Content appears on the right part of the window shown in Figure 1. One of the features of VPen system is multimedia annotation tool that was available to students to annotate learning materials by underlining, highlighting and adding geometric figures, Figure 1. Students use the comment box provided by VPen system to add text annotation. The comment box also allows students to add multimedia content, such as picture or audio into text annotations. Besides, we designed recording mechanism in our annotation system so students are able to record their speech and add it as an audio file into textual annotation. In order to listen to audio file which is added into text annotation students have to click an icon of audio file, Figure 1. After students login into VPen system individual mode to work with learning materials was presented by default; this is when student is able to add and see own annotations only. In group mode student besides adding and seeing own annotations is able to see peer’s annotations as well. This feature allows students to work in groups and interact. To switch from individual mode into group mode student have to check group mode option of VPen system, left part of Figure 1. Peer’s annotation is presented as an annotation that includes peer’s name while own annotation has no name. Figure 1 is a screen capture from one of students interface and its right part presents an example of students group work when two students annotated learning material. One student, the

Effect of Multimedia Annotation System on Improving English Writing

5

owner of interface, added geometric figures, in this case a rectangle and attached two text annotations, one with textual content and one with audio file; another student, the peer, added text annotation. Another feature of grouping mechanism is Member list which is a floating menu that shows all names of class members and the names are ordered by students’ contributions - the quantity of annotation; students with higher quantity of annotations are the first on the list.

Fig. 1. VPen multimedia annotation system

3.3 Learning Activities The following learning activities were designed in our study: (1)Activity A Individual picture sharing. In this learning activity teacher prepared learning material which included text and pictures related to topic students learn. After reading the contents of the learning material, students were asked to find a picture from Internet related to the topic and add it to the activity web site. Students were also assigned to describe a picture by adding annotation with textual content. (2)Activity B Commenting on peer work. In this learning activity students were asked to share their work from activity A (picture and textual description) with peers. Students were divided into pairs and each student was assigned to review partner’s work. After reviewing, students were asked to give their opinion or ask question with relation to peers’ works by adding annotations with audio content. (3)Activity C Questioning and answering. This learning activity was divided into two parts. In the first part, students were provided with 5 different pictures related to subject and they had to compose at least 3 questions to each picture. In the second part, students shared their questions with peers; each student had to listen to other student’s questions and answer them. Questions and answers were added as annotation with audio to activity web site. (4)Activity D Complete the story. In this learning activity students were supposed to work individually. Each student were assigned five different pictures where only

6

W.-Y. Hwang, R. Shadiev, and S.-M. Huang

first picture was provided with the text – the beginning of the story, so student had to read commencement of the story and complete the story based on the rest of the pictures by adding annotation with textual content to activity web site. (5)Activity E Story Relay. In this learning activity, students were assigned to create a story based on another five pictures provided by teacher. Students were divided into small groups with 2-3 students in each, then they discussed the sequence of the pictures and outline of the story among group members, and afterward, students took turn to record the contents of the story. (6)Activity F Pictures and their description. During this learning activity students had to choose 3-5 pictures from database related to sport competition that has been hold at school a week earlier and add them to activity web site. Students added textual descriptions to the pictures first and then oral description of the sport event. (7)Activity G Opinion exchange. This learning activity is a sequel of learning activity F where students were asked to share their work from learning activity F with peers. Then students were assigned to review peers pictures and read or listen descriptions to the pictures and school event. After that students had to add their comments, suggestion or questions as annotations with textual or audio content.

4 Data Analysis and Results 4.1 Questionnaire Analysis The questionnaire was conducted after experiment to understand students’ perception and attitude toward the VPen system and learning English activities. Out of twenty students, twenty six valid answer sheets were received. Questionnaire contains fourty seven questions in five dimensions. Obtained Cronbach α values were higher than .80 in all dimensions which according to Wortzel [28] showed that the results reached high reliability. In Perceived easy to use VPen system (the degree to which a student believes that using VPen system would be free from effort) and Perceived usefulness of VPen system (the degree to which a student believes that using VPen system would enhance his/her performance during learning activities) dimensions, all items were ranked with high score. Most of the students agreed that it was easy to use VPen system and that VPen system was useful during learning English activities. In Perceived usefulness of activities (the degree to which a student believes that online English learning activities would enhance his/her learning achievement) dimension, almost all items were ranked high showing that students in general thought that learning English activities were useful in improving writing and speaking English skills. In interview students mentioned that the pictures in the activities could facilitate their imagination and help them create text or audio, thus they think the activity was useful and fun. Only three items with relation to practice speaking skills were ranked with low score; item “By sharing my opinions to my partners in the activities, like B, F and G, I can improve my English speaking ability” ranked as 3.5 out of 5, item “In general, sharing ideas in the activities, like B, F and G, can assist my English speaking” ranked as 3.6 out of 5 and item “In general, I think activities C and E are helpful for my English

Effect of Multimedia Annotation System on Improving English Writing

7

speaking practice” ranked as 3.6 out of 5. Interview with the students and teacher’s observation revealed that students thought interaction in those activities happened in asynchronous mode which wasn’t useful to practice speaking skills. For example, when students recorded description of the picture and share it, students couldn’t get immediate feedback from the peers on what is incorrect in their speech, like content, pronunciation, structure or others; students had to wait until their peers listen to their recording and then give a feedback. This could be done during the same class or during next class which is two weeks away. So students just kept speaking and recording their speech without awareness of mistakes. In Perceived playfulness of activities (the degree to which a student believes that online English learning activities would fulfill students’ intrinsic motives) dimension, five items were ranked high showing that some students thought that proposed learning English activities were playful. In interview students mentioned they felt interested in proposed by experiment creative way to develop their writing or speaking skills instead of following standard curricula where students supposed to give expected correct answers or use test. However, seven items in this dimension were ranked low (3.2-3.5 out of 5). In item “Doing the online English activities I always pay attention on it” sixteen students choose undecided which lowered the rank of the item. The possible reason is that it was students’ first experience on working in such new environment and on such learning activities; some students couldn’t understand certain parts of learning materials, some couldn’t understand what they were asked to do during activities since no hint or Chinese translation were provided. So some students could be distracted from participation at the learning activities by asking questions from teacher or peers or they were asked questions by peers. The item “I can experience an authentic English learning environment when doing the English online activities” was scored low by some students since we couldn’t provide instant feedback to their text or audio content. In the future study we will try to overcome this and abovementioned shortcomings. The rest of the items were scored low since some students were not familiar with activity design; it was productive oriented where students had to create text or audio content, so students found participating at activities as difficult and they feel dislike. In Perceived attitude of using VPen (the variable that is jointly determined by perceived easy to use, perceived usefulness, perceived usefulness of activities and perceived playfulness of activities) dimension, almost all items were ranked high showing that the students had positive attitude to use VPen in general. Only item “I frequently use VPen system to do English learning activities” was ranked low, as 3.5 out of 5. From interview with the students we found that all students used VPen system during class time only; the students had no time to use VPen after class since they were busy with studying other subjects and completing homework and assignments. Besides, students were not allowed by parents to use computer after class; the parents were afraid that students spend most of the time playing computer games instead of studying and doing homework. 4.2 Correlational Analyses Students’ attitude toward VPen use. We analyzed relationship between ease of VPen use, usefulness of VPen, usefulness of learning English activities, playfulness

8

W.-Y. Hwang, R. Shadiev, and S.-M. Huang

of learning English activities and students’ attitude toward VPen use. Person correlation was employed and the result of analysis showed that all four dimensions have significant correlation with students’ attitude toward VPen use (Table 1). From questionnaire and interview with the students analysis we found that majority of students think that VPen system was easy to use, it was also useful during learning English activities and students perceived that designed learning English activities were playful and useful in improving writing and speaking skill. Thus, students’ beliefs about the system and the activities resulted to students’ positive attitude toward using VPen. Table 1. Correlational analysis of students’ attitude toward VPen use Ease of VPen Use Attitude toward VPen .606(**) use *pt(2)0.01. There exist significant differences between post-test and pre-test. It proves that intervention has significant effects than non-interventions. Table 3. Results of T-test: the first baseline condition (pre-test) and second baseline conditions (post-test) Paired Differences Std. Mean Deviation

Pair 2 Pretest-posttest -1.333

1.155

Std. Error Mean

.667

95% Confidence Interval of the Difference

Lower

-4.202

Sig. (2-

t df tailed) 1.535 -2.000 2 .184

Upper

We know from t-Table that, t(df)0.05=t(2)0.05=4.303, sample t=2.000t(2)0.01. There exist significant differences between post-test and pre-test. It proves that the performance has significantly increased after import the intervention once again.

4 Conclusions and Discussion After analyze the experimental results, we found it matches with our assumption: After introduce four inter-school collaborative learning activities to pupils aging 11 and 12, they had significant performance improvement on conceptual change and cognitive behavior. The intervention of “inter-school collaboration” and “attribute keyword’s extraction”, dramatically improve collaborative conceptual change and promote performance while building complicated conceptions. Plus, the result of the quasi-experiment shows that there are functional relations between applying and drawing intervention in sequence and performance of students’ conceptual change. This could be explained by heterogeneous group members’

32

X. Xu and Y. Ren

cognitive conflict caused by exchanging information process. The conflicts arouse students’ curiosity; hence stimulate their motivation for deeper exploration. Later they complement others’ knowledge while discuss difference between areas to find properties and connecting rules of these properties. With the guide from their teacher and supports from “Keyword Extraction” tool, students generalize and summarize properties of temperature and heat. They found 3 properties for temperature: “Measureable”, “Value”, and “degree of cold and hot”; 4 for Heat: “energy”, “radioactivity”, “conductibility”, “perceivable” (it is inaccurate in nature science, but enough for elementary science). Our research shows senior pupils distinguish the two concepts by summarizing their properties. During this progress they understand the differences in essence and build new scientific conceptions. We learn that key to design learning activities is differential learning theory, which would arouse cognitive conflict. For example, in active 1: Students’ feeling differs when they felt the water in 25°C, in the vary day outdoor temperature in Haikou and Karamay are +27°C and -17°C. So when student exchange information about their feeling towards the water online, they would encounter problems, thus led to cognitive confliction. “Problem” above means they felt water at same time and in the same temperature, but their feeling differs due to different environment. This made them curious for a reason, and will guide students into discussion for one. Later they realized “temperature is scalable and measurable, but heat could be perceived” all by themselves. In the exact same way, students learned more properties within the next 3 activities, to a final understanding of the difference between temperature and heat. Processes mentioned above follows the rules of conceptual learning in psychology. In actual research we designed those four comparative experimental learning activities under them. We hope learners could easily get those attributes as we have highlighted them, and we could see from research results that this design had achieved our purposes and relative effects. Considering the potential deficiencies in this research, we have to run some future investigations about scientific conceptions’ maintain that learners keep in different situations. For example, would learners always sustain same understanding in different environment if they already distinguish conceptions of temperature and head in school? More importantly: Would learners keep adopting scientific conceptions in daily life while interacting with phenomenon related to those conceptions? Meanwhile, what aspect or clue makes learners keep using scientific conceptions instead of misconception? In addition, compare Internet supported inter-school collaborative method to traditional one with instructors’ oral introduction on different conceptions [13]. Which is the more effective one in helping learners learning “Conceptual distinguish of Temperature and Heat” so that they induce conceptual changes and conceptual transfers? What would be the reasons and clues for conceptual changes and transfers? Those would be future researches we plan to deal with.

References [1] Wandersee, J.H., Mintzes, J.J., Novak, J.D.: Research on alternative conceptions in science. In: Gabel, D.L. (ed.) Handbook of Research on Science Teaching and Learning, pp. 177–210. Macmillan, New York (1994)

Computer-Supported Collaborative Conceptual Change

33

[2] Sozbilir, M.A.: Review of Selected Literature on Students’ Misconceptions of Heat and Temperature. Bogazici University Journal of Education 20(1), 25–40 (2003) [3] DiSessa, A.A.: A History of Conceptual Change Research. In: The Cambridge HandBook of The Learning Sciences, pp. 256–281. Cambridge University Press, NY (2006) [4] DiSessa, A.A.: A History of Conceptual Change Research: Threads and Fault Lines. In: The Cambridge HandBook of The Learning Sciences, pp. 265–282. Cambridge University Press, NY (2006) [5] Glynn, Yeany, R.H., Britton, B.K.: A Constructive view of learning science. In: The psychology of learning science, pp. 205–217 (1991) [6] White, R., Gunstone, R.: Probing Understanding. The Falmer Press, London (1992) [7] Black, J.B., McClintock, R.O.: An interpretation construction approach to constructivist design. In: Constructivist Learning Environments: case studies in instructional design, pp. 25–32. Educational Technology Publications, Englewood Cliffs (1996) [8] Levin, I., Druyan, S.: When Sociocognitive Transaction Among Peers Fails: The Case of Misconceptions in Science. Child Development 64, 1571–1591 (1993) [9] Hennessey, M.G.: Students’ ideas about their conceptualization: Their elicitation through instruction. Paper presented at the annual meeting of the National Association for Research in Science Teaching, Atlanta, GA (1993) [10] Wisniewski, E.J., Medin, D.L.: On the interaction of theory and data in concept learning. Cognitive Science (18), 221–281 (1994) [11] Bruner, J.S., Goodnow, J.J., Austin, G.A.: A study of thinking. Wiley, New York (1956) [12] Robert, H., Hala, A.: A content analytic comparison of F2F and ALN case-study discussion [DB/OL] (September 10, 2005), http://www.hicss.hawaii.edu/HICSS36/HICSSpapers/CLALN01.pdf [13] Johnson, D.W., Johnson, R.T.: Implementing cooperative learning. Contemporary Education 63(3), 173–180 (1992) [14] Scardamalia, M., Bereiter, C.: Higher Levels of Agency for Children in Knowledge Building: A Challenge for the Design of New Knowledge Media. Learning Sciences 1(1), 37–68 (1991)

Optimization Technique for Commercial Mobile MMORPG* Wang Jianmin1, Zheng Zibin2, Peter Tam3, and Liu Jianping1 1

School of Information Secience &Technology ,Sun Yat-sen University, Guangzhou, China 2 The Chinese University of Hong Kong, Hong Kong, China 3 CTO, Gameislive Corporation, 5/f, 4 Victory Avenue, Kowloon, Hong Kong [email protected], [email protected], [email protected], [email protected]

Abstract. In this paper, we discuss the model and strategy to improve the performance of mobile MMORPG (Massively Multiplayer Online Role-Playing Game). For mobile games, there are lot restrictions like the JAR, Heap and Bandwidth on mobile devices. Memory optimization and network bandwidth reduction are crucial in mobile MMORPG. In this paper, we discuss our optimization experience in crafting a commercially launched game “ZhanGuo” [14], which is proven to run in more than 800 types of mobile phones owned by a million subscribers since 2005. Based on our first-hand experience, we have found out three interesting properties in mobile online game development: 1) extremely small memory factor in both storage and runtime heap memory; 2) the extensive use of graphics and background maps in game; 3) frequent and massive network communication. The idea we presented is only the very first attempt to tackle the issues in the domain of mobile online game, which has a huge opportunity for future development. In addition, we present some open problems in the domain of mobile MMORPG deserving further research and exploration. Keywords: MMORPG, optimization, mobile phone.

1 Introduction Nowadays, mobile devices are becoming powerful enough to tackle an ever-widening range of applications. And wireless networking is becoming ubiquitous, cheap, and low-power. Mobile games are becoming more and more popular. Datamonitor predicts that wireless gaming will generate US $17.5 billion in annual revenue worldwide. Constrained by the computing ability, memory and screen size of mobile * Supported by the National Natural Science Foundation of China under Grant No. 60776096; the National High-Tech Research and Development Plan of China under Grant Nos. 2007AA01Z236; the Natural Science Foundation of Guangdong Province of China under Grant No. 9151027501000035, and Guangdong Province scientific and technological project under Grant No. 009B010800017. X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 34–45, 2010. © Springer-Verlag Berlin Heidelberg 2010

Optimization Technique for Commercial Mobile MMORPG

35

phones, developing mobile MMORPG meet a lot of challenges. Performance optimization technologies, such as storage memory and heap memory conservation technologies and network communication traffic reduction technologies, are very crucial to cope with these challenges. J2ME (Java 2 Micro Edition) is a platform for mobile devices. The Mobile Information Device Profile (MIDP) 1.0 and 2.0 are two key elements of J2ME, which provides a standard Java runtime environment for mobile information devices. There are three types of memory in J2ME applications, namely: Storage Memory. Storage memory is a persistent and write-protected memory, which used to store the JAR file of the applications. The storage memory is usually very limited in mobile phones. Minimizing the JAR size of the applications can conserve the storage memory, as well as reduce the application download time over-the-air. Stack and Heap. The run time data of J2ME applications are stored in stack and heap memory. The primitivetype variables, such as int, long, boolean are stored in stack. And the runtime objects are saved in the heap memory. Unfortunately, heap sizes of the majoring mobile devices are extremely limited ranging from 200K to 2048K bytes. Consuming all the available heap memory will cause “OutOfMemory Exception”. Some strategies to measure and save the heap memory are discussed in this paper. Persistent Storage. Persistent Storage is a non-volatile place for storing data. J2ME applications use the Record Management System (RMS) to utilize the persistent storage. In this paper, we present an innovative way of using persistent storage as a cache to conserve the heap memory and enhance the performance of the game. In this paper, we will discuss some general strategies to optimize the memory and network performance of the mobile MMORPG. The performance of the game “ZhanGuo” has been upgraded by using these strategies. In Section 2, the related works are introduced. In section 3, technologies to shrink the JAR size are presented. In section 4, the heap memory measurement and optimization technologies are discussed. In section 5, strategies for optimizing mobile communication bandwidth are provided. Finally we concluded the result the present future work in section 6

2 Related Work There are many studies on Java class files compression and packing algorithms. In 1997, Ernst et al. [4] describe measurements conservation about Java code compression. In 1998, Nenil et al. [10] propose a compression scheme, which operates on individual class files. Bradley et al. [10] introduce the Jazz file format to store Java programs, which can compress the class files to a degree that far exceeds what is possible with a JAR file. In 1999, Pugh [13] developed a wire-code format for collections of Java class files, which can achieve 1/2 to 1/5 of the size of the corresponding compressed JAR file. In 2001, William et al. [12] describes the design and implementation of a method for producing compact, bytecoded instruction sets and interpreters for them. In 2002, Frank et al. [2] explore extraction techniques such as the removal of unreachable methods and redundant fields, for reducing application size. In our work, we use technologies that is independent the JAR format, such that our result is compatibility to the standard JAR format as well as other special format.

36

J. Wang et al.

The researches on Java heap memory conservation are active. In 2003, G. Chen et al. [3] propose a set of memory management strategies to reduce heap footprint of embedded Java applications that execute under severe memory constraints. Shaylor et al. [6] describes a JVM™ architecture designed for very small devices. In 2004, Kato et al. [5] propose a Java computation mechanism with a de/compression module that reduces Java heap memory demand. In our work, we use a different angle to solve the runtime heap memory limitation by using persistent storage as a cache, which is analogy to virtual memory concept in modern operating system. About the network performance, in 1995, Ramon Caceres et al. [9] proposed a scheme to lessen the effects of non congestion related losses over wireless networks. In 1996, Edward W. Knightly et al. [1] use the Hybrid Bounding Interval Dependent (H-BIND) to provide end-to-end statistical performance guarantees to bursty traffic. In 2001, Mark D. Corner et al. [4] present a system that copes with the challenges of providing interactive video for mobile devices through a division along time scales of adaptation. In our work, some strategies to reduce the mobile network communication traffic of the mobile MMORPG are presented.

3 Jar Size Optimization In this section, some JAR size optimization technologies are discussed. By using these technologies, we reduce the JAR size of the leading mobile MMORPG in China: “ZhanGuo” from 130,649 bytes to 64,791 bytes. Because of the high compress rate, the game is small enough to run on almost all Java-enabled phones in the market. 3.1 Combine Small Image into Big Image In a mobile MMORPG, graphics occupy a large space of the JAR file. The most common approach to reduce the graphics size is combining small PNG (Portable Network Graphic Format) images into a single bigger image. By reducing the header of PNG files, a lot of space can be saved. For small images, the overhead of PNG header is obvious; on bigger images, the saving are not as dramatic [11]. For example, in our game, there are 6 types of avatars (3 male, 3female), each avatar consists of 4 directions, and each direction consists of 3 animation frames. So there are 6 ×4 ×3=72 images for the avatars in the game. Originally, these images, which are 16×24pixel PNG files, separately stored as c1.png, c2.png…..c72.png in the JAR file. To reduce the JAR size, 12 small animation images of the same avatar are combined into a single bigger image. See Figure 1, the sizes of these 12 small PNG image files before combined are as follow (Bytes): 711, 728, 844, 745, 676, 743, 838, 759, 693, 719, 850, 766 bytes. After combined, the size is greatly reduced.

Fig. 1. The Combined Image

Optimization Technique for Commercial Mobile MMORPG

37

12

sizeBeforeCombined = ∑ fileSize(i ) = 9027bytes i =1

sizeAfterCombined = 3315bytes compressRate = 3315 / 9072 = 36.5% In the game, only the desired part of the combined avatar image is drawn to the screen when used, the following codes are used to draw the desired part of the image: int playerOffset=(dir + (animation x > y’s lower limit }, which describes the time cost in the learner’s learning of a knowledge point is in the personalized time range of the learner’s learning of the corresponding knowledge domain. y is the hasLessTime: hasLessTime={ | x is a KDPointInhabit individual corresponding individual of InhabitActive y’s lower limit > x }, which describes the time cost in the learner’s learning of a knowledge point is less than the lower limit of the personalized time range of the learner’s learning of the corresponding knowledge domain. hasMoreFrequency: hasMoreFrequency={< x , y > | x is a KDPointQA individual y is the corresponding individual of QAActive x > y’s upper limit}, which describes the number of questions in the learner’s learning of a knowledge point is more than the upper limit of the personalized number range of questions during the learner’s learning process. y is hasInFrequency: hasInFrequency={< x, y > | x is a KDPointQA individual the corresponding individual of QAActive y’s upper limit > x > y’s lower limit }, which describes the number of questions in the learner’s learning of a knowledge point is in the learner’s personalized number range of questions during the learner’s learning process.

∧

∧

∧

∧

∧

∧

∧

∧

∧

Knowledge Preference Based Learning Community Construction

127

∧

hasLessFrequency: hasLessFrequency={< x, y > | x is a KDPointQA individual y is the corresponding individual of QAActive y’s lower limit > x }, which describes the number of questions in the learner’s learning of a knowledge point is less than the lower limit of the personalized number range of questions during the learner’s learning process. hasMoreNumber: hasMoreNumber={< x, y > | x is a KDPointPractice individual y is the corresponding individual of PracticeActive x > y’s upper limit }, which describes the correct rate of exercises in the learner’s learning of a knowledge point is more than the upper limit of the personalized correct rate range of exercises during the leaner’s learning process. y is hasInNumber: hasInNumber={< x, y > | x is a KDPointPractice individual the corresponding individual of PracticeActive y’s upper limit > x > y’s lower limit}, which describes the correct rate of exercises in the learner’s learning of a knowledge point is in the personalized correct rate range of exercises during the leaner’s learning process. hasLessNumber: hasLessNumber={< x, y > | x is a KDPointPractice individual y is the corresponding individual of PracticeActive y’s lower limit > x }, which describes the correct rate of exercises in the learner’s learning of a knowledge point is less than the lower limit of the personalized correct rate range of exercises during the leaner’s learning process.

∧

∧

∧

∧

∧

∧

∧

According to the concepts and relations described above, many situations can be judged such as whether the learner has spent more time learning a knowledge point, whether the learner has fewer questions in learning a knowledge point. The learner’s personalized information will be described according to the learner’s actual learning situation. By using the concepts and their relations described above, the learner’s knowledge preference on the knowledge point of the knowledge domain can be obtained or modified according to the learner’s instant learning situations. 2.2 Reasoning of Knowledge Preference By the knowledge preference model defined above, the relationship among knowledge domain, knowledge preference and the learning situation can be reasoned. The concrete reasoning rules are as follows: In order to express the rules conveniently, some identifiers are defined here: p denotes one Learner individual, k denotes one KnowledgeDomain individual, kdp denotes one KDPoint individual which is included in k, kdpi denotes the corresponding KDPointInhbit individual of p and kdp, ia denotes the corresponding InhabitActive individual of p and k, kdpp denotes the corresponding KDPointPractice individual of p and kdp, pa denotes the corresponding PracticeActive individual of p and k, kdpq denotes the corresponding KDPointQA individual of p and kdp, qa expresses the corresponding QAAvctive individual of p and k, t denotes the corresponding Test individual of k, tm denotes the corresponding TestMark individual of p and t, ts denotes the corresponding TestStandard individual of t.

128

H. Deng, Y. Zhan, and Q. Mao

1) The reasoning rules of knowledge preference of the learner The reasoning rule of the learner having more preference on the knowledge point: ∈ hasMoreTime∧ ∈ hasMoreFrequency∧ ∈ hasMoreNumber p has more preference on kdp

If the learner’s three indicators are all more than the upper limit of the corresponding personalized ranges, we think the learner has more preference on the knowledge point. The reasoning rule of the learner having medium preference on the knowledge point:

( ∈ hasMoreTime∧ ∈ hasMoreFrequency∧ ∈ hasInNumber) ∨ ( ∈ hasInTime∧ ∈ hasMoreFrequency∧ ∈ hasMoreNumber) ∨ ( ∈ hasMoreTime∧ ∈ hasInFrequency∧ ∈ hasMoreNumber) p has medium preference on kdp If the learner’s two of the three indicators exceed the upper limit of the corresponding personalized ranges, and the rest is in the corresponding personalized range, we think the learner has medium preference on the knowledge point. The reasoning rule of the learner having less preference on the knowledge point: ( ∈ hasInTime∧ ∈ hasInFrequency∧ ∈ hasMoreNumber) ∨ ( ∈ hasMoreTime∧ ∈ hasInFrequency∧ ∈ hasInNumber) ∨ ( ∈ hasInTime∧ ∈ hasMoreFrequency∧ ∈ hasInNumber) p has less preference on kdp

If the learner’s two of the three indicators are both in the corresponding personalized ranges, the rest exceeds the upper limit of the corresponding personalized range, we think the learner has less preference on the knowledge point. 2) The reasoning rules of knowledge preference in the learner’s further learning As the learner’s learning situations are not static, the knowledge preference should be modified when the learner’s learning situation has changed. Therefore, it is necessary to make some modification when the learner’s learning situation has changed. The reasoning rule of changing the learner’s preference to more preference: ( ∈ hasMoreTime∧ ∈ hasMoreFrequency∧ ∈ hasMoreNumber) ∧( ∈ hasInPrefer∨ ∈ hasLessPrefer) p's preference on kdp should be changed from medium or less preference to more preference

If the learner’s three indicators all exceed the upper limit of the corresponding personalized ranges, and the learner has already had medium or less preference on the knowledge point before, we think the learner has more preference on the knowledge point. The reasoning rules of changing learner’s preference to medium preference:

Knowledge Preference Based Learning Community Construction

129

(( ∈ hasInTime∧ ∈ hasMoreFrequency∧ ∈ hasMoreNumber) ∨ ( ∈ hasMoreTime∧ ∈ hasInFrequency∧ ∈ hasMoreNumber) ∨ ( ∈ hasMoreTime∧ ∈ hasMoreFrequency∧ ∈ hasInNumber)) ∧ ∈ hasLessPrefer p's preference on kdp should be changed from less preference to medium preference

If the learner’s two of the three indicators both exceed the upper limit of the corresponding personalized ranges, the rest is in the corresponding personalized range, and the learner has less preference on the knowledge point before, we think the learner has medium preference on the knowledge point. (( ∈ hasInTime∧ ∈ hasInFrequency∧ ∈ hasMoreNumber) ∨ ( ∈ hasMoreTime∧ ∈ hasInFrequency∧ ∈ hasInNumber) ∨ ( ∈ hasInTime∧ ∈ hasMoreFrequency∧ ∈ hasInNumber)) ∧ ∈ hasMorePrefer p's preference on kdp should be changed from more preference to medium preference If the learner’s two of the three indicators are both in the corresponding personalized ranges, the rest exceeds the upper limit of the corresponding personalized range, and the learner has already had more preference on the knowledge point before, we think the learner has medium preference on the knowledge point. The reasoning rule of changing learner’s preference to less preference: ( ∈ hasInTime∧ ∈ hasInFrequency∧ ∈ hasInNumber) ∧( ∈ hasMorePrefer ∨ ∈ hasInPrefer) p's preference on kdp should

be changed from more or medium preference to less preference If the learner’s three indicators are all in the corresponding personalized ranges, and the learner has already had more or medium preference on the knowledge point before, we think the learner has less preference on the knowledge point. The reasoning rules of deleting the learner’s preference: ( ∈ hasLessTime∨ ∈ hasLessFrequency∨ ∈ hasLessNumber) ∧( ∈ hasMorePrefer∨ ∈ hasInPrefer∨ ∈ hasLessPrefer) p's preference on kdp should be deleted If the learner’s three indicators are all lower than the lower limit of the corresponding personalized ranges, and the learner has already had more or medium or less preference on the knowledge point, we think the learner has no preference on the knowledge point. 3) The reasoning rules of the learner’s cognition situation The reasoning rule of the learner having good cognitive situation: ∈ hasStudied ∧ ∈ hasGoodMark p accomplishes k's learning objective great

If the learner finishes the knowledge domain and gets a good test mark, we think the learner accomplishes the knowledge domain great.

130

H. Deng, Y. Zhan, and Q. Mao

The reasoning rule of the learner having normal cognitive situation: ∈ hasStudied ∧ ∈ hasNormalMark p accomplishes k's learning objective normally

If the learner finishes the knowledge domain and gets a normal test mark, we think the learner accomplishes the knowledge domain normally. The reasoning rule of the learner having bad cognitive situation: ∈ hasStudied ∧ ∈ hasBadMark p accomplishes k's learning objective badly

If the learner finishes the knowledge domain and gets a bad test mark, we think the learner accomplishes the knowledge domain badly. By using the reasoning rules mentioned above, the learner’s various learning situation and knowledge preference can be obtained in our system. With the help of knowledge preference model, it is convenient to construct a personalized learning community for the learner.

3 Learning Community Construction Based on Knowledge Preference According to the knowledge preference model mentioned above, it is easy to get the learner’s knowledge preference in his or her studying the knowledge domain. The learner’s knowledge preference can be used as the basis of the construction of the learning community for learning the knowledge domain. There are two steps for learning community construction. Step1: Quantification. The learner’s knowledge preference of the knowledge domain will be quantified as the learner’s knowledge preference vector of the knowledge domain. Step2: Clustering. According to the knowledge preference vector, learners will be divided into several groups by using the clustering algorithm. And the groups are the learning communities constructed for learners. 3.1 Quantification of the Learner’s Knowledge Preference

According to the learning situation of the learner’s studying the knowledge domain, his or her knowledge preference of the knowledge domain is available with the help of the knowledge preference model. Here, we give a preference value correspondence with the each preference. For instance, 1.0 represents the learner having more preference on the knowledge point, 0.7 represents the learner having medium preference on the knowledge point, 0.4 represents the learner having less preference on the knowledge point, and 0 represents the learner having no preference on the knowledge point. Then, the learner’s knowledge preference vector of the knowledge domain is available. For instance, we denote U= {x1, x2 … xm} as a knowledge domain which includes m knowledge points, xi as a knowledge point and m as the number of knowledge points.

Knowledge Preference Based Learning Community Construction

131

Then, we can use the following vector to describe the learner’s preference, which is l= (l1, l2 … lm). Each attribute of l represents the learner’s preference of the corresponding knowledge point included in the knowledge domain. So if there is a set of learners’ knowledge preference of the knowledge domain, that is S= {s1, s2 … sn}, and each learner’s knowledge preference is described as a vector like l, the corresponding reactor matrix of S is available, that is S= (rij) n×m. Then rij describes the learner si’s preference value on the knowledge point xj. 3.2 Algorithm Procedure of Clustering on the Basis of the Learner’s Knowledge Preference

Due to existence of many vague concepts which are cannot be described precisely, fuzzy clustering methods are used in our system to represent data structure of attribute accurately. The concrete steps of the algorithm as follows: Step1: Choose an appropriate function, f: S × S → [0.1] ‚calculate qij = f ( xi, xj ) what denotes the degree of similarity between xi and xj, and build the corresponding similarity matrix Q = ( r i j ) n × n . Step2: Calculate Q 2 = Q × Q, Q 4 = Q 2 × Q 2 ,..., step by step until Q2k = Qk .Then Q is k

fuzzy equivalence matrix and denoted as Q * . Step3: Select each value in Q * , if the value is bigger than threshold λ , it will be set as 1; otherwise, it will be set as 0. And then we get a corresponding *

λ -cut fuzzy matrix,

-1

Q = (a1, a2… an) . If ai = aj, then si and sj are put into the same learning community; λ

otherwise, ai and aj are separately put into different learning communities.

4 Service Support Based on Learning Community When the learner accomplishes a knowledge domain, the system will analyze his or her learning condition and generate a personalized learning community for him or she by using the algorithm described in section 3. Once the learner’s condition changes, the learning community will change too. In the learning community, learners can talk with other members to share their learning experience or propose their confusion during the learning process. And if the learner accomplishes a knowledge domain badly, an excellent learner will be selected to help him or her by talking. The concrete rule of recommending learner is as follows: p1 accomplishes k badly ∧ p2 accomplishes k greatly ∧ p1 and p2 are in the same learning community p2 is the learner recommended to talk with p1

132

H. Deng, Y. Zhan, and Q. Mao

If the leaner accomplishes the knowledge domain badly, an excellent learner which is in the learner’s learning community will be recommended to help him or her study the knowledge domain. Furthermore, the excellent learner’s learning track will be showed to the learner. According to the knowledge preference model in E-Learning environment, by using the Stanford University’s Protégé ontology modeling tool [6], the concepts and relations described in section 2 are constructed. Semantic Web’s OWL language documents are formed finally. An ontology-based learning community with knowledge preference is developed by using Jena reasoning equipment [7]. Some subjects such as “Java”, “C++” and “The fundament of database” have been available in our system now. And the system has been tested by the students of School of Computer Science and Telecommunication engineering of Jiangsu University till now. Take the learner s5 for example. In order to simplify the example, three kinds of individuals are constructed by the system. They are: one Learner individual: s5(090805, s5, 23, male, [email protected]), one KnowledgeDomain individual: KD1(KD1, Fundamentals of database system, KD1_document, KD1_objectove), one TestMark individual: (090801, KD1, 85). Figure 2 shows the learner s5’s knowledge preference and learning community when he accomplishes the knowledge domain “Fundamentals of database system” which includes four knowledge points: “database summary”, “database system”, “

Fig. 2. The example shows the learner s5’s knowledge preference after his accomplishment of studying the knowledge domain “Fundamentals of database”. And the members of his learning community are showed too.

Knowledge Preference Based Learning Community Construction

133

database management database”, and “relational database management system”. From the Figure 2, it is found that s5 has more preference on “database summary” and “ relational database management system”, less preference on “database system” and no preference on “database management system” in his learning of the knowledge domain “Fundamentals of database”. In addition, the members of s5’s learning community are described on the web site. There are five people in s5’s learning community including s3, s4, s5, s6 and s7. It is found that all of them have similar knowledge preference after they has studied the knowledge domain “Fundamentals of database system”. If the learner s5 learns “Fundamentals of database system” and accomplishes the learning objective badly, the website like Figure 3 will be shown to s5. From the website, an excellent learner s4 is recommended to help s5 study “Fundamentals of database system”. Meanwhile, the learning track of s4 will be shown to s5 as a reference.

Fig. 3. The example shows the learner s5’s knowledge preference after his bad accomplishment of studying the knowledge domain “Fundamentals of database”. In addition of show of the member of his learning community, an excellent learner s4 is recommended to help s5.

According to the result of the questionnaires of the students who used the system, we can see that the student’s learning interest is stimulated greatly and most students say their loneliness is declined by talking with the members of their own learning communities and their learning efficiency has been raised greatly. And most students are willing to study with the system.

134

H. Deng, Y. Zhan, and Q. Mao

5 Conclusion According to the characteristics of the E-Learning environment, by making good use of the ontology’s characteristic, the learning situations are described reasonably and efficiently. Therefore, a knowledge preference ontology model in the E-Learning environment is presented in this paper. According to the knowledge preference rules described above, the learner’s knowledge preference caused by different learning situations can be concluded. The corresponding learning communities are formed according to the different knowledge preference and the learner’s loneliness is declined by talking with the members of the learning community. Therefore, the learner’s learning interest would be stimulated more effectively and the learner’s learning efficiency would be raised greatly. Acknowledgements. This work is supported partly by National Natural Science Foundation of China under Grant No. 60273040 and Open Foundation of State Key Lab for CAD& CG, Zhejiang University.

References 1. Li, H., He, Q., Wu, Z.: The Research on Personalized Study System Based on Individual Student in Web. Journal of Computer Engineering and Applications 38(13), 239–242 (2002) (in Chinese) 2. Tu, C., Corry, M.: E-learning communities. Quarterly reviews of distance education 3(2), 207–218 (2002) 3. Gascue, J.M., Fernández-Caballero, A., González, P.: Domain Ontology for Personalized E-Learning in Educational Systems. In: Proceedings of the Sixth International Conference on Advanced Learning Technologies, pp. 456–458. IEEE Press, Los Alamitos (2006) 4. Matheus, C.J., Kokar, M., Baclawski, K.: A Core Ontology for Situation Awareness. In: Proceedings of the 6th International Conference on Information Fusion, pp. 545–552. IEEE Press, San Francisco (2003) 5. Zhan, Y., Wang, J., Mao, Q.: Nested Knowledge Space Model and Awareness Processing in a Collaborative Learning Environment. Journal of Computer Research and Development 42(7), 1159–1165 (2005) (in Chinese) 6. Protégé, http://protege.stanford.edu 7. JENA, http://jena.sourceforge.net

Developing an Online History Educational System to Present the Progression of Spatial Regions Jia-Jiunn Lo and Hsiao-Han Tu Department of Information Management, Chung-Hua University, HsinChu, Taiwan [email protected], [email protected]

Abstract. This research developed an online history educational System, named HES-SPATO2. It is unique in that it integrates the indispensable elements of history events such as person, space, and time for increasing the understandability of complicated history learning materials. HES-SPATO2 employed GIS concept of information layers and web-based technology to help students acquire skills for historical thinking. In HES-SPATO2, history learning objects can be formed by integrating the elements of history events to specify “who” initiated the history event, “whom” the event influenced, “what” they did, “when” the event happened, and “where” the event happened. HES-SPATO2 applied temporal logic to reason the temporal relationships between history events. It also applied spatial logic to reason the spatial relationships between regions. Combining temporal logic and spatial logic makes it possible to present not only movements of persons and explanation texts but also the progression of spatial regions along the history events in animation defined in each SPATO (Spatial, Person, Action/Attribute, and Temporal Object) and therefore makes the learning material more understandable. This system not only brings us the obvious advantage of anywhere-anytime learning but also provides ways to revitalize the teaching and learning of history. Keywords: History educational system, GIS, temporal logic, spatial logic.

1 Introduction The idea of e-learning, in the Internet era, allows for flexibility of access from anywhere and usually at anytime. Essentially, “it allows participants to collapse time and space” [1] (p.4). It also facilitates challenging activities that enable learners to link new information to old and acquire meaningful knowledge [2]. In fact, online courses are an increasingly common feature in education worldwide. Many of these courses include history courses. However, the challenge for history teachers and course developers is to construct a learning environment that is constructive, interactive, multisensory, and studentcentered. History lessons should provide learning tools to help students develop abilities and strategies to think historically and to contextualize their interpretation of events in terms of when and where the events happened. Although online learning has become popular in the delivery of history materials, most of the history learning X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 135–144, 2010. © Springer-Verlag Berlin Heidelberg 2010

136

J.-J. Lo and H.-H. Tu

material that is available online is heavily text-based and teacher-centered. For instance, [2] addressed that most online learning materials are created by simply translating traditional hardcopy textbooks into electronic formats, which does not authentically facilitate effective online learning. [3] also pointed out that most web pages created for history teaching “tend to be text-heavy and to contain few opportunities for innovative interactions by the user beyond a set of links to explore”, and this tendency presents challenges when creating online history courses. [4] reminded that education is not only about access to content. The greatest affordance of the web for educational use should be the profound and multifaceted increase in communication and interaction capability that it provides. Since history is highly related to geographical location, researchers tend to include geographic information in history educational systems and develop systems based on GIS (Geographic Information System) [5,6]. A GIS map is composed of many separate, overlapping information layers, each of which has its own meaning [7]. By applying these information layers, temporal and spatial attributes are integrated in spatialtemporal coordinates. There are several advantages in using GIS to develop history educational systems. Firstly, it is useful for spatial information integration. Then, through the bridging of time and space, and by efficiently gathering, saving, editing, managing, analyzing and displaying all spatial information, it can successfully integrate various types of spatial data and digital information management systems, and further demonstrate the values of data from different perspectives [5,6]. In addition, these systems can store, retrieve, map, and analyze geographic information about history and, hence, increase the understandability of the learning materials. [8] cautioned that the visual elements incorporated into history materials should be very simple and should convey only one point at a time, because students tend to lose their concentration on the overall flow of the lecture when they are faced with a complicated image. His suggestion is in accordance with the claims of [4]. According to the theory of cognitive learning, because humans have limited short-term memory capacity, information should be grouped into meaningful sequences so that the students are not overwhelmed by too much information. In our system, the GIS concept of information layers is employed for spatial information integration. In other words, temporal and spatial attributes of people, events, and objects are integrated. Students can study not only maps, but all information related to them. In addition, the feature size can be adjusted. Students can zoom in to see them close up. Different features can be visualized with different information layers. With this interface, while students can view courses with integrated person, space, and time with high understandability of learning materials. A space-time data model can help us understand the dynamic processes of history. Modeling of space and time attempt to define the real life phenomena through objects as well as their relationships and constraints [9]. [9] used cell complexes for representing spatial-temporal objects. This approach presents space, time, and spatialtemporal data at the same time. In their research, reality is perceived as an object called Spatio-Temporal-Attribute Object (STAO) for modeling purposes. STAO is useful to recode the spatial data through time. Person is indispensable and the most important element for history events. Not only should the spatial and temporal components be included but also the person component. Though successfully applied

Developing an Online History Educational System

137

to spatial-temporal data models as in urban applications, it is difficult for STAO to model history events which are centered around persons. Therefore, it is suggested to extend STAO by including “person” for modeling history events [10,11]. History events happened in sequences. Though the systems using the GIS technology of information layers can present historical data for different times, the dynamics of history events are not clearly presented. Without presenting history events step by step makes it difficult for students to understand the relationships among history events. Therefore, it is crucial to develop a history educational system that can present the dynamics of history events in sequences. In the previous research, we developed a spatial-person-temporal online history educational system named HES-SPATO (History Educational System based on SPATO) [10,11]. This system was based on SPATO (Spatial, Person, Action/Attribute, and Temporal Object). It was developed based on the Sharable Content Object Reference Model (SCORM). SCORM promotes efforts to create flexible learning materials with reusability, durability, accessibility, and interoperability. SPATO, the backbone of the proposed system, is an object to specify “where”, “who/whom”, “what”, and “when” about history events. By integrating SPATOs with different types of assets, SCORMbased history learning materials such as SCOs, lessons, and courses can be formed accordingly. With application of the temporal logic [12] to reason the temporal relationships between history events, HES-SPATO makes it possible to present history events in sequences with animation to clearly illustrate the dynamics of history events. In addition, the GIS concept of information layers is applied to develop the learning module. With the developed learning module, students can select features, displayed with different layer, to realize history events more clearly, in the manner they want, zoom in to see features at closer range, view variety supplement learning materials at the same time, and view SCOs according to his own progress and steps. HES-SPATO proposed innovative ways to transform traditional static resources into online modules that maximize student interaction with history instructional materials by developing of a data model to integrate the person, space, action/attribute, and time information of history events, applying the GIS concept of information layers to develop the learning module, and applying the temporal logic to reason the temporal relationships between history events. However, as [13] mentioned, history refers not only to what happened in the past but also to the account of the past events, situations, and processes. It represents accounts of multilayered and multifaceted human experiences across time and space. Thinking historically is a challenging task for students because they must not only follow events across time, but across categories of analysis. Usually, changes of political spatial regions, such as founded, split, mergence, authority change, and rename, happened along with history events across time and it is essential to understand such spatial region changes for learning history. To this end, this research developed HES-SPATO2, an extended version of HES-SPATO by applying the spatial logic RCC-8 [14,15] to reason the progression of spatial regions along history events.

138

J.-J. Lo and H.-H. Tu

2 HES-SPATO2 Architecture Being an extended version of HES-SPATO, similarly, HES-SPATO2 is based on SPATO as developed by [10,11]. HES-SPATO2 has eleven components based on SCORM and the learning module as illustrated in Fig. 1.

Asset Pool_Other

SCO Base

Lesson Base

Spatial Reasoning

Sequencing Constructor

Sequencing Template

Manifest Transformer

Manifest Template

SPATO

Asset Pool_SPA

Course Base

Learning Module

Fig. 1. System Architecture of HES-SPATO2

In HES-SPATO2, there are two types of Asset Pools, Asset Pool_SPA and Asset Pool_Other. Asset Pool stores assets developed by learning material designers or imported from other Internet sources. An asset is the smallest physical unit in SCORM. It can be any electronic format that can be delivered to a web client. Asset Pool_SPA stores the Space, Person, and Action/Attribute assets which are required to define SPATO. On the other hand, Asset Pool_Other stores all other types of assets which are not used to define SPATO. A SPATO is an object to specify “where”, “who/whom”, “what”, and “when” about history events [10,11]. It specifies indispensable elements of history events such as “where” the event happened, “who” initiated it, “whom” it influenced, “what” they did, and “when” it happened. The SPATO is a logical object rather than a physical file. It includes pointers for linking “Space”, “Person”, and “Action/Attribute” assets to form a learning object for presenting the elements of a history event by integrating assets stored in the Asset Pool_SPA. History events can be presented in animation along the historical timeline when these assets are integrated with the “Time” component specified in SPATO. SPATO also provides the basis for spatial reasoning. A SPATO is defined as: . ObjectID

Developing an Online History Educational System

139

is the unique feature identifying a SPATO. SpaceClass specifies the geographical location where the event took place. PersonClass_actor specifies the person who initiated the event and PersonClass_acceptor specifies the person whom it influenced. Action/AttributeClass specifies what the event was. TimeClass_start specifies when the event started and TimeClass_end specifies when it ended. For example, for the event “Mary fought with John at school from 9am to 10am.” The SPATO is defined as . Usually, progressions of political spatial regions happened along with history events across time include founded, split, mergence, authority change, and rename. It is beneficial to understand such spatial region changes for learning history. Spatial Reasoning applies RCC-8, the spatial logic proposed by [14,15], to present the progression of spatial regions along history events. SCO Base stores learning objects used in the system. The associated “Space”, “Person”, and “Action/Attribute” assets, stored in the Asset Pool_SPA, are integrated into SPATO to form a history learning object. All learning objects follow the SCORM standard. A SCO can exist alone and is the smallest logical unit in HESSPATO. Lesson Base is used to manage lessons combined by SCOs in the SCO Base. A lesson may include one or more SCOs. All lessons belong to SCORM content aggregations. Courseware designers create a lesson by selecting and organizing SCOs in the SCO Base. A course is composed of two elements, lessons and sequential relationships among lessons. Lessons are defined in Lesson Base and the sequential relationships are defined in Sequencing Constructor. In Sequencing Constructor, courseware designers define the relationships among lessons by sequencing objects (SO) which are provided by Sequencing Template. After defining the relationships among lessons, the layout of a course is defined in Manifest Transformer. Sequencing Template provides SOs to the Sequencing Constructor for designing the relationships among lessons. In HES-SPATO, two types of sequencing templates, forward and choice, are provided. A manifest is a document that contains a structured inventory of the content of a package. In Manifest Transformer, the layout of each lesson is defined by using the template provided by Manifest Template. After designing the layout of a lesson, HES-SPATO will translate the course structure into a Web file and store the file in Course Base. Manifest Template provides manifest templates to Manifest Transformer for designing the layout of a lesson. Course Base stores and manages course files. Students study history by navigating the course files stored in Course Base through the learning module.

3 Learning Module of HES-SPATO2 For a long time, people have studied history with models such as maps. In the last decades, it has become possible to put these models inside computers. These computer models for maps make up a GIS. In a GIS, students can study not just maps,

140

J.-J. Lo and H.-H. Tu

but all possible information related to maps. With the right data, students can see whatever they want in whatever part of the world interests them [16]. A GIS map is composed of many separate, overlapping information layers, each of which has its own features [7]. Using GIS technology is useful for integrating spatial information. Applying such concept of information layers, temporal and spatial attributes of people, events, and objects are integrated in spatial-temporal coordinates. In HESSPATO2, the GIS concepts are applied to develop the learning module so that students can view courses integrated with person, space, and time. Fig. 2 illustrates the HES-SPATO2 learning module.

Fig. 2. Learning Module of HES-SPATO2

(1) Time line: Presenting the start time, end time, and current time of the event. (2) Event title: The name of the current history event. (3) Map: Displaying the movement of persons and progression of political spatial regions along the timeline (4) Person: Participants in the history event that moved along the timeline according to the historical path. (5) Real-time information: A piece of short text to describe the key content of the current animation step of a history event. (6) Map view controller: Any part of the map can be viewed with different resolution levels by using panning and zooming tools. (7) Animation controller: Students may let the system show the animation automatically or step by step. (8) Information layer controller: Students can choose different information layers as they want. (9) Supplementary data: All related supplementary data are shown in this frame. Learners can freely select different types of supplementary materials.

Developing an Online History Educational System

141

The GIS concepts have been applied to develop the learning module of HESSPATO2. Students can choose features from different information layers. The system presents not only movements of persons and explanation texts [10,11] but also the progression of spatial regions along with history events with the temporal relationships according to the TimeClass defined in each SPATO. Usually, changes of political spatial regions, which include founded, split, mergence, authority change, and rename, happened along with history events across time and it is essential to understand such spatial region change for learning history. Founded, split, and mergence are related to changes of spatial shape and size. On the other hand, for authority change and rename, only properties of the regions are changed without changing the spatial shape and size. Taking the progression of Nan County (Nan Jun) during the Three Kingdoms period of Chinese history as an example. The Three Kingdoms were made up of three kingdoms-Wei, in northern China, Shu to the southwest, and Wu in the southeast. The leaders of the kingdoms strove to reunite the empire and were therefore at constant warfare. Fig. 3 demonstrates the progression of Nan County along the timeline. Political region

Split

Nan County (South)

(Wu)

Nan County (North)

(Wei)

Nan County Xiangyang County

(Han)

(Shu) (Wei)

(Wu)

(Wei)

Linjiang County Yidu County

Mergence Rename Authority change (Authority)

(Wei)

(Wu)

(Shu)

Aug/A.D. 208 A.D. 209 Dec/A.D. 208 A.D. 210

Time

Fig. 3. Progression of Nan County (Nan Jun) during the Three Kingdoms Period of Chinese History

Based on the spatial logic adopted from RCC-8 [14,15], Fig. 4 – Fig. 8 illustrate the progression of such changes of political spatial regions displayed in the learning module. Authorities are differentiated with different colors. Nan County was founded by Han Dynasty in 106 B.C. (Fig. 4). Then its authority was changed to Wei and it was splitted into three counties, Nan, Xiangyang, and Linjiang in August, A.D. 208 (Fig. 5). In December, A.D. 208, it was further splitted into Nan County (South) whose authority was changed to Wu and Nan County (North). At the same time, the authority of Linjiang County was also changed to Wu (Fig. 6). In A.D. 209, Nan County (South) and Nan County (North) were merged and the authority was changed to Wu (Fig. 7). In A.D. 210, authorities of Linjiang County and Nan County were hanged to Shu and Linjiang County was renamed as Yidu County (Fig. 8).

142

J.-J. Lo and H.-H. Tu

Nan County (founded in 106 BC by Han Dynasty)

Fig. 4. Nan County (Before August, A.D. 208)

(a) (b) (c)

Fig. 5. Nan County (August, A.D. 208): (a) Xiangyang County - splitted from Nan County and authority changed to Wei in August, A.D. 208; (b) Linjiang County- splitted from Nan County and authority changed to Wei in August, A.D. 208; (c) Nan County - splitted from Nan County and authority changed to Wei in August, A.D. 208

(a) (b) (c)

Fig. 6. Nan County (December, A.D. 208): (a) Linjiang County - authority changed to Wu in Dec/AD 208; (b) Nan County (North) - splitted from Nan County in Dec/AD 208; (c) Nan County (South) - splitted from Nan County and authority changed to Wu in December, A.D. 208

Developing an Online History Educational System

143

(a)

Fig. 7. Nan County (A.D. 209): (a) Nan County - merged from Nan County (South) and Nan County (North); authority changed to Wu in A.D. 209

(a) (b)

Fig. 8. Nan County (A.D. 210): (a) Yidu County - renamed from Linjiang County and authority changed to Shu in A.D. 210; (b) Nan County - authority changed to Shu in A.D. 210

4 Conclusions Many researchers call on historians and history educators to restructure historical study and historical teaching to take advantage of the potential of computer technology to think and present historical information and ideas visually. Our research motivated us to use computer technology to model historical thinking. The proposed system is unique in that it integrates the indispensable elements of history events such as person, space, and time for increasing the understandability of complicated history learning materials. It employed GIS concept of information layers and web-based technology to help students acquire these skills for historical thinking. In HES-SPATO2, history learning objects can be formed by integrating SPATOs with the associated “Space”, “Person”, and “Action/Attribute” assets, stored in the Asset Pool_SPA. SPATO specifies “who” initiated the history event, “whom” the event influenced, “what” they did, “when” the event happened, and “where” the event happened. With SPATO, a history educational system with sharable learning objects that are integrated with spatial, person, and temporal history information was implemented. HES-SPATO2 applied temporal logic to reason the temporal relationships between history events. HES-SPATO2 also applied spatial logic to reason the spatial relationships between regions. Combining temporal logic and spatial logic makes it possible to present not only movements of persons and explanation texts but also the progression of spatial regions along with history events in animation defined in each SPATO and therefore makes the learning material more understandable. This system not only brings us the obvious advantage of anywhere-anytime learning but also provides ways to revitalize the teaching and learning of history.

144

J.-J. Lo and H.-H. Tu

Acknowledgments. This research is supported by National Science Council of Taiwan, R.O.C. (Grant No: NSC 95-2520-S-216 -002-MY3) and Chung-Hua University, Taiwan, R.O.C. (Grant No: CHU-95-2520-S-216-002-MY3).

References 1. Ally, M.: Foundations of Educational Theory for Online Learning. In: Anderson, T., Elloumi, F. (eds.) Theory and Practice of Online Learning. Athabasca University, Canada (2004) 2. Bonk, C.J., Reynolds, T.H.: Learner-centered Web instruction for higher-order thinking, teamwork, and apprenticeship. In: Khan, B.H. (ed.) Web-based Instruction. Educational Technology Publications, Englewood Cliffs (1997) 3. Vess, D.: History in the Digital Age: A Study of the Impact of Interactive Resources on Student Learning. The History Teacher 37(3), 385–399 (2004) 4. Anderson, T.: Toward a Theory of Online Learning. In: Anderson, T., Elloumi, F. (eds.) Theory and Practice of Online Learning. Athabasca University, Canada (2004) 5. Academic Sinica, Introduction to Chinese and Taiwan Historical GIS, http://ccts.ascc.net/download/presentation_20050138_en.pdf (retrieved January 1, 2009) 6. Lo, F.-J.: Literary and Geographic Space-Time Information System Design and Application: Take Sushi’s Poems for Example. In: The Second Taipei International Conference on Digital Earth, Taipei, Taiwan, R.O.C (2004) 7. Haag, S., Cummings, M., McCubbrey, D.J.: Management Information Systems: For the Information Age. McGraw-Hill Irwin Companies, Inc., New York (2004) 8. Coohill, J.: Images and the History Lecture: Teaching the History Channel Generation. The History Teacher 39(4), 455–465 (2006) 9. Raza, A.: Object-Oriented Temporal GIS for Urban Applications. Ph.D. Dissertation, INF, University of Twente, The Netherlands (2001) 10. Lo, J.-J., Chang, C.-J., Tu, H.-H., Yeh, S.-W.: Applying GIS to Develop a Web-Based Spatial-Person-Temporal History Educational System. Computers & Education 53(1), 155–168 (2009) 11. Lo, J.-J., Chang, C.-J., Tu, H.-H., Yeh, S.-W.: HES-SPATO: An Online History Educational System Based on SCORM. In: Pan, Z., Cheok, A.D., Müller, W., Rhalibi, A.E. (eds.) Transactions on Edutainment II. LNCS, vol. 5660, pp. 160–175. Springer, Heidelberg (2009) 12. Allen, J.: Maintaining Knowledge about Temporal Intervals. Communications of the ACM 26(11), 832–843 (1983) 13. Yilmaz, K.: A Vision of History Teaching and Learning: Thoughts on History Education in Secondary Schools. The High School Journal, 37–46 (December 2008/January 2009) 14. Cohn, A.G., Bennett, B., Gooday, J., Gotts, N.M.: Qualitative Spatial Representation and Reasoning with the Region Connection Calculus. GeoInformatica 1(1), 1–44 (1997) 15. Randell, D.A., Cui, Z., Cohn, A.G.: A Spatial Logic based on Regions and Connection. In: 3rd International Conference on Knowledge Representation and Reasoning (1992) 16. Ormsby, T., Napoleon, E., Burke, R., Groessl, C., Feaster, L.: Getting to Know ArcGIS Desktop: Basics of ArcView, ArcEditor, and ArcInfo. ESRI Press, Redlands (2004)

A Bibliometric Study of E-Learning Literature on SSCI Database Johannes K. Chiang, Chen-Wo Kuo, and Yu-Hsiang Yang Department of Management Information System National Chengchi University Taipei, Taiwan 11623 [email protected], [email protected], [email protected]

Abstract. This paper investigates the publishing trends of e-learning literature catalogued in SSCI database during 1967-2009. Our findings indicate that (1) the quantity of recent research on e-learning is expanding remarkably; (2) the frequency indexes of authors productivity do not appear to abide by Lotca’s Law; (3) most research papers on e-learning are generated by multiple authorship; and (4) applications of e-learning have most found in research areas such as Education & Educational Research, Information Science & Library Science, and Computer Science/Interdisciplinary Applications. Finally, future directions of research on e-learning are considered. Moreover, according to Bradford’s Law, the three zone ratio comparisons almost equal as 1 8 82, which means the data does match Bradford’s Law. And the seven core journals in e-learning are identified and analyzed.

：：

Keywords: E-learning, Lotka, Bradford, author productivity, bibliometrics.

1 Introduction Since the first scholarly paper in electronic-learning, or e-learning, appeared in 19671, and according to SSCI database explorations of the possibilities of e-learning have seen a vigorous development, especially in the last fifteen years owing to the Information and Communication Technology (ICT). Kruse (2004) analyzed developments in ICT-based E-learning through 1996 to 2002 and argued that e-learning started from 1996 and reached its peak in 2000. Around 2002, it returned to previous level. ICTbased E-learning leads the development of knowledge management to the community of Management and Business Administration. (Huang and Yang, 2009; Garcia, 2009) Researches that provide practicable e-learning options to narrow the digital divide are also discussed from the viewpoint of sociology and social science. (Friedman and Deek, 2003; Laschewski, 2008; Mutula, 2008). Aiming to arrive at a better understanding of the quantitative characteristics of recorded information such as the research institutions and subject areas of e-learning literatures on social science, this paper employs a bibliometric methodology towards a 1

According to SSCI database, we found one retrieved record “[Anon] (1967), New Electronic Learning Center. Educational Technology 7(3): 16-17.” The author is anonymous.

X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 145–155, 2010. © Springer-Verlag Berlin Heidelberg 2010

146

J.K. Chiang, C.-W. Kuo, and Y.-H. Yang

literature review concerning its patterns of productivity and publishing trends. Last but not least, applying Lotka’s law and Bradford’s law to analyze author productivity and core journals in this field (within 1967 and 2009) respectively, will lead to discovery in literature features and research tendency in the future.

2 Description of E-Learning The term e-learning has several definitions. Some refer to e-learning as either packaged content pieces or technical infrastructures; some consider it to be asynchronous to autonomous learning; while still others view e-learning as a synchrony for collaborative learning. It is generally agreed, however, that e-learning is an approach to learning. The e-learning group of National Center for Supercomputing Applications (NCSA) provides a general definition: According to Rosenberg (2001), E-learning refers to the use of Internet technologies to deliver a broad array of solutions that enhance knowledge and performance. It is based on three fundamental criteria: 1. E-learning is networked, which makes it capable of instant updating, storage / retrieval, distribution and sharing of instruction or information. 2. It is delivered to the end user via a computer using a standard Internet technology. 3. It focuses on the broadest view of learning–learning solutions that go beyond the traditional paradigms of training. Chute, Thompson & Hancock (1999) point out that e-learning has the following characteristics, for which it will become an indispensable vehicle of education: „ „ „ „ „ „ „

A fast, efficient dissemination to all areas of learning methods. High benefit-cost ratio, because they can provide higher-quality courses and to reduce travel. For busier people, if enough incentives and will enhance the rate of course. Information or knowledge acquisitions are the latest, but also the use of a faster-to-work. Courses can be shorter or longer periods, for learners to provide a more flexible and diversified. To increase the number of learning rather than being limited by space and cost. To consult experts in the field, and more, and quickly find the answer.

American Society for Training Development (ASTD) defines e-learning as the intermediate of transferring information through electronic means to assist learning. In brief, e-learning involves the use of technologies to enhance learning. Keegan (1990) defines e-learning as a teaching method whereby the teacher employs ICT, digital learning materials, along with pedagogical theories, to facilitate and support learning. E-learning may also be considered as a learning activity, which is based on on-line communication. Learners can browse through the materials, conduct discussions, perform in tests, or do exercises after logging onto the system. The definition of e-learning in the national program is defined as an instructional and learning approach, which makes it possible for students to learn better and for teachers to teach more effectively, in contrast with traditional classroom settings. In

A Bibliometric Study of E-Learning Literature on SSCI Database

147

general, e-learning has several advantages, including: (1) it reduces the costs of teaching and training in organizations such as schools, government departments, businesses, and private institutions; (2) it stimulates learners' motivations to learning; (3) it provides opportunities for interactive and flexible learning (Hwang, 2003). On account of these benefits of e-learning, an increasing number of countries allocate funds and resources to encourage research in e-learning. In Europe, the Commission of the European Communities (2001) published the guidelines of an e-learning policy. The E-learning Action Plan-Designing tomorrow’s education; The Secretary of Commerce in the United States (2002) made a report entitled "2020 VisionsTransforming Education and Training through Advanced Technologies". The government of Taiwan (2002) also initiated a five-year e-learning project called National Science and Technology Program and allocated four billion NT dollars to promote e-learning in Taiwan.

3 Research Findings and Discussion Our research utilizes the data from the Social Sciences Citation Index (SSCI) of Web of Science created by the Institute for Scientific Information (ISI). An empirical retrieval method was operated using "e-learning", "distance learning", "digital learning",

Fig. 1. Number of published papers in e-learning

Fig. 2. Citation in each year (Source: SSCI database)

148

J.K. Chiang, C.-W. Kuo, and Y.-H. Yang

Fig. 3. Distribution of Top 10 productive country

or "electronic learning" to retrieve data. A total of 1,944 papers in e-learning published during 1967-2009 were found. Figure 1 and 2 indicate the growth of the number of academic papers and their annual citations. The results appear to suggest that the number of papers about e-learning has distinctively increased since 1995, and that citations of papers in e-learning are also on the increase each year. It appears that e-learning has received much attention from researchers, which leads to a rapid growth of related papers and citations, as illustrated in Figure 1 and 2. According to the numerical data, a large amount of research papers published during 2004-2009 have been catalogued in the SSCI database, with the distribution rate of 163(9.12%), 157(8.78%), 227(12.7%), 311(16.72%), and 254(13.07%) respectively, against the total number of papers indexed. With regard to the distribution of nationalities of the 1,632 papers indexed for this research, top-rated nations with the most publications catalogued in SSCI database during 1967-2009 are elicited, as illustrated in Figure 3. According to the statistics, the United States outnumbers all the other nations in terms of number of papers, with a total of 591 papers (30.4%) retrieved. It is then followed by England (340 papers; 17.49%), Taiwan (139 papers; 7.15%), Australia (66 papers; 3.4%), and Canada (61 papers; 3.14%). Table 1. The top ranking institutions with record counts greater than or equal twelve

Institution Name Open Univ Natl Cheng Kung Univ Univ Sheffield Univ N Carolina Natl Chiao Tung Univ Natl Cent Univ Univ Illinois Univ S Africa Open Univ Netherlands Univ Pretoria

Count 79 19 19 18 17 15 13 13 12 12

% 4.06% 0.98% 0.98% 0.93% 0.87% 0.77% 0.67% 0.67% 0.62% 0.62%

Country England Taiwan England USA Taiwan Taiwan USA South Africa Netherlands South Africa

A Bibliometric Study of E-Learning Literature on SSCI Database

149

Table 1 offers a closer look at the distribution of academic institutions by which the indexed papers were submitted. It is observed that, among the 13 institutions whose record counts of indexed papers are greater than nine, 9 are located in USA, followed by England and Germany. Based on the published studies related to e-learning, USA is the most productive country. Taiwan is also among the top ten. (25, 3.15%). And its institution National Chengchi University ranks No.4, holds 12 published papers as that of Harvard University (USA), Santa Fe Institution (USA) and University Groningen (Netherlands). Table 2 offer an investigation into the authors who have written more than six papers related to e-learning research. The top 3 authors are Richardson, J.T.E. (13, England), Barker, P (12, England), Deeson, E (10, England), and Koper, R (10, Netherlands). Table 2 shows that the research in computer science and Education related to e-learning are the mainstream. With a view to provide insights into the future directions of e-learning research, the discussion now turns to the applications of electronic learning. Table 3 shows the top 10 subject areas in which e-learning are most widely utilized based on our retrieval runs of SSCI database. Among all the subject matters listed here, education & educational research take the lead with 878 papers (45.16%) against the total of 1,944papers retrieved. Information science & library science ensues, with 301 papers recorded (15.48%). This is then followed by Computer Science/Interdisciplinary Applications, Table 2. The top ranking of published papers in e-learning based on authors

Author

Count %

Compri sing % Country of country

Institution

Subject area

Richardson, JTE Barker, P Deeson, E Koper, R Chen, CM Huang, YM Tseng, SS Amandi, A Barron, T Chen, GD

13 0.67% 2.20% England

Open Univ

Education

12 10 10 9 7 7 6 6 6

Univ Teesside Blackwell Publ Ltd Open Univ Natl Chengchi Univ Natl Cheng Kung Uni Natl Chiao Tung Univ Unicen Univ Univ Rochester Natl Cent Univ

Info Sci & Lib Sci Education Education Info Sci & Lib Sci Eng Sci Comp Sci Comp Sci Comp Sci Comp Sci

Du Preez, M

6

Univ S Africa

Info Sci & Lib Sci

Galagan, PA

6

0.31% 1.02% USA

Liaw, SS Price, B

6 6

0.31% 4.32% 0.31% 1.02%

Sun, PC

6

0.31% 4.32%

Tattersall, C

6

0.31% 9.84%

0.62% 0.51% 0.51% 0.46% 0.36% 0.36% 0.31% 0.36% 0.31%

2.03% 1.69% 16.4% 6.47% 5.04% 5.04% 75% 1.02% 4.32%

England England Netherlands Taiwan Taiwan Taiwan Argentina USA Taiwan South 0.31% 14.3% Africa

Amer Soc Training Business Development Taiwan China Med Univ Education England RCN Inst Info Sci & Lib Sci Natl Kaohsiung Normal Comp Sci Taiwan Univ Netherlands Open Univ Education

150

J.K. Chiang, C.-W. Kuo, and Y.-H. Yang Table 3. The top ranking of published papers in e-learning based on subject areas

Rank Subject Area

Count

%

1

Education & Educational Research

878

45.16%

2

Information Science & Library Science

301

15.48%

3

Computer Science/Interdisciplinary Applications

176

9.05%

4

Computer Science/Information Systems

138

7.10%

5

Nursing

111

5.71%

6

Psychology, Multidisciplinary

79

4.06%

7

Business

75

3.86%

8

Management

61

3.14%

9

Education, Scientific Disciplines

54

2.78%

10

Communication

50

2.57%

Fig. 4. Yearly Distribution of Top 5 Subject Areas

with 176 papers (9.05%) related to e-learning. Referring to Figure 4, Education & Educational Research is growing fast after 2000, while Information Science & Library Science and Computer Science/Interdisciplinary Applications are gradually growing.

A Bibliometric Study of E-Learning Literature on SSCI Database

151

To summarize, a vigorous development in e-learning literature is observed, based on our retrieval run over the SSCI database. The top five nations with the most papers in e-learning are the US, England, Taiwan, Canada, and Netherlands. In addition to our bibliometric data, which suggest the quantitative growth of research in e-learning, efforts have been made to further promote the development of this emerging field. The United Nations, for instance, has called for a massive investment in e-learning from governments, especially developed countries – on the ground that such efforts may contribute to a better understanding of nature, society, and related areas of interests, which will then empower the knowledge of the mankind in the near future.

4 Bradford’s Law and Journal Literature Samuel C. Bradford in 1934 introduced Bradford’s Law which is a pattern to estimates the exponentially diminishing returns of extending a search for references in Table 4. The distribution of E-Learning journals

No. No. of of articles journal (A) (B) 159 1 152 1 131 1 (A) Core 58 2 41 1 33 1 30 1 29 1 27 1 23 1 21 2 19 1 17 2 16 1 (B) 15 1 Relevant 12 3 11 4 10 4 9 10 8 2 7 5 6 13 5 13 4 23 (C) 3 41 marginal 2 85 1 296

Accumulated Journals (C) 1 2 3 5 6 7 8 9 10 11 13 14 16 17 18 21 25 29 39 41 46 59 72 95 136 221 517

(D) = (A)*(B) 159 152 131 116 41 33 30 29 27 23 42 19 34 16 15 36 44 40 90 16 35 78 65 92 123 170 296

(E) = Accumulated (D) 151 303 434 550 591 624 654 683 710 733 775 794 828 844 859 895 939 979 1069 1085 1120 1198 1263 1355 1478 1648 1944

Log (acc. Journals) 2.1790 2.4814 2.6375 2.7404 2.7716 2.7952 2.8156 2.8344 2.8513 2.8651 2.8893 2.8998 2.9180 2.9263 2.9340 2.9518 2.9727 2.9908 3.0290 3.0354 3.0492 3.0785 3.1014 3.1319 3.1697 3.2170 3.2887

152

J.K. Chiang, C.-W. Kuo, and Y.-H. Yang

science journals. The law principle impose a formulation that if journals in a field are sorted by number of articles into three groups, each group approximate to one-third of all articles, then the number of journals in each group will be proportional to 1:n: n². (Tsai, 2003) The 1,944 publish papers in this study distributed in 517 journals. Table 4 provides the number of publish paper each journal and other information ranking by the number of publish paper according to the zoning of Bradford Law. Table 5 also provides the ratio comparisons of 3 zones, that is ratio of published paper each zone of zone A, B, C, 7:53:458. It almost equal to 7:56:448 as 1 8 82. That is, A: B: C = 1: n: n². The result matches the explanations of Bradford Law. Table 6 shows the core journals in E-Learning. The number one journal, British Journal of Educational Technology has 159 published papers (8.18%), greater than the number of publish paper of number two journal, Computers & Education (152, 7.82%). It is also observed that the main subject areas of the five journals are Education and Education Technology.

：：

5 Lotka’s Law and Author Productivity Lotka's law of scientific productivity of authors is a good example with respect to such empirical laws. Lotka deduced an inverse square law relating the authors of Table 5. The literature brief distribution in E-Learning based on journal

A B C

(1) No. of journal 7 52 458

(2) No. of articles 24 376 1041

(3) Range of No. of articles 33~159 6~30 1~5

(4) Average articles 3 7 2

Table 6. The seven core journal titles and their statistics in E-Learning

Title

Count

British Journal Of Educational Technology

159

Computers & Education

152

Educational Technology & Society

131

Journal Of Computer Assisted Learning

58

Training & Development

58

Electronic Library

41

Etr&D-Educational Technology Research And Development

33

% Acc. % 8.18 8.70 % % 7.82 16.5 % 2% 6.74 23.2 % 6% 2.98 26.2 % 4% 2.98 29.2 % 2% 2.11 31.3 % 3% 1.70 33.0 % 3%

A Bibliometric Study of E-Learning Literature on SSCI Database

153

Table 7. Author distribution of Lotka’s Law

count

% of author(s)

Accumulated % of Author(s) Sn (X)

Expected % of author(s)

Accumulated Expected % of author(s) Fo (X)

Absolute Value |Fo (X)-Sn (X)|

1 2 3 4 5 6 7 9 10 12 13

0.8858 0.0816 0.0213 0.0057 0.0014 0.0024 0.0005 0.0003 0.0005 0.0003 0.0003

0.8858 0.9674 0.9887 0.9944 0.9957 0.9981 0.9987 0.9990 0.9995 0.9998 1.0000

0.8651 0.0888 0.0235 0.0091 0.0044 0.0024 0.0015 0.0006 0.0005 0.0002 0.0002

0.8651 0.9539 0.9774 0.9865 0.9909 0.9933 0.9948 0.9954 0.9959 0.9961 0.9963

0.0207 0.0134 0.0113 0.0078 0.0048 0.0048 0.0039 0.0035 0.0036 0.0037 0.0037

Record

published papers to the amount of papers written by each author in 1926. The data represented in the decennial index of Chemical Abstracts specifically and the Auerbach's Geschichtstafeln der Physik as the name index, Lotka plots the number of authors against the number of contributions contributed by each author on a logarithmic scale. Lotka dictated these points are closely scattered around a straight line having a depth slope of approximately negative two. This empirical observation as Lotka concludes provided the following equation (Chung and Cox, 1990). an = a1/nc, n= 1, 2, 3,..,

(1)

Where an = the number of authors publishing n papers, a1 = the number of authors publishing one paper, and c = a constant. (in Lotka’s case, c = 2) Taking the log of both sides of (1), we obtain log(an) = log(a1) - clog(n).

(2)

In the computation of the "best empirical value", the constant c for data related to elearning by fitting a line to the empirical frequency distribution. The regression results show that c = 3.28. If the estimated a1 is 0.8651, then the equation (1) will be stated as follows:

an = 0.8651 / n3.28 It is possible to check whether e-learning literature matches the Lotka’s Law by K-S statistical test. According to K-S test, as demonstrated in Table 7, as if Dmax=0.0407 and the sampling number is bigger than 35, then the threshold value will be 1.63/37031/2 = 0.02679, while the number of accumulated authors will be 3703.

154

J.K. Chiang, C.-W. Kuo, and Y.-H. Yang

Although the fact that Dmax is less than the threshold value, the result matched the generalized Lotka’s law, which indicates that the Lotka’s law is author productivity distribution data related to e-learning literature.

6 Conclusion E-learning is one of fastest growing field of research in recent years. Having analyzed the characteristics of literature on e-learning as well as author productivity distribution, one may expect that the number of research papers in this area will continue to increase. The main research institutions which have yielded the most publications of e-learning papers are located in the US, England, and Taiwan. Our research findings also suggest that there is a centralization tendency of institution distribution in both England and Taiwan. The frequency indexes of author productivity distribution abide by Lotka’s Law. Applications of e-learning are most active in the fields of Education & Educational Research, Information Science & Library Science, and Computer Science/Interdisciplinary Applications. It is also observed in our research that e-learning papers are usually generated by multiple authorships. Moreover, according to Bradford’s Law, the three zone ratio comparisons almost equal as 1 8 82, that means the data does match Bradford’s Law. And, the seven core journals in E-Learning are identified and analyzed. The effects of perceived issues for the potential of e-learning are worthy to study further.

References 1. American Society for Training Development, http://www.astd.org/ 2. Chung, K.H., Cox, R.A.K.: Patterns of Productivity in the Finance Literature: A Study of the Bibliometric Distributions. The Journal of Finance 45(1), 301–309 (1990) 3. Chute, A.G., Thompson, M.M., Hancock, B.W.: The McGraw-Hill handbook of distance learning. McGraw-Hill, New York (1999) 4. Commission of the European Communities. The E-learning Action Plan- Designing tomorrow’s education (March 28, 2001), http://ec.europa.eu/education/archive/elearning/annex_en.pdf 5. Friedman, R.S., Deek, F.P.: Innovation, and education in the digital age: Reconciling the roles of pedagogy, technology, and the business of learning. IEEE Transactions on Engineering Management 50(4), 403–412 (2003) 6. Garcia, B.C.: Developing Connectivity: a PKM path for higher education workplace learners. Online Information Review 33(2), 276–297 (2009) 7. Huang, S.L., Yang, C.W.: Designing a semantic bliki system to support different types of knowledge and adaptive learning. Computers & Education 53(3), 701–712 (2009) 8. Hwang, R.H.: Web-based Distance Learning. The Minister of Economic, Taipei (2003) 9. Keegan, D.: The foundations of distance education. St. Martins Press, New York (1990) 10. Kruse, K.: The State of e-Learning: Looking at History with the Technology Hype Cycle (September 14, 2004), http://www.e-learningguru.com/articles/hype1_1.htm 11. Laschewski, L.: Boundless Opportunities? - Visions and Setbacks to Digital Learning in Rural Areas. Eastern European Countryside 14, 79–91 (2008) 12. Lotka, A.J.: The frequency distribution of scientific productivity. Journal of the Washington Academy of Sciences 16, 317–323 (1926)

A Bibliometric Study of E-Learning Literature on SSCI Database

155

13. Mutula, S.M.: Evolving Paradigms in the Networked world and their Implications for Information Management in African Libraries. African Journal of Library Archives and Information Science 18(2), 89–102 (2008) 14. National Science & Technology Program for E-learning. The Blueprint of the National Science & Technology Program for E-Learning in Taiwan (June 15, 2002) (in Chinese) 15. Rosenberg, M.J.: E-learning: Strategies for Delivering Knowledge in the Digital Age. McGraw-Hill, New York (2001) 16. Tsay, M.Y.: The Characteristic of Informetrics and Bibliometrics. Hwa-Tai bookstore, Taipei (2003) (in Chinese) 17. Tsay, M.Y., Jou, S.J., Ma, S.S.: A Bibliometric Study of Semiconductor Literature, 19781997. Scientometrics 49(3), 491–509 (2000) 18. The Secretary of Commerce. Visions 2020 - Transforming Education and Training through Advanced Technologies (2002), http://isites.harvard.edu/fs/docs/icb.topic443490.files/ 2020visions.pdf 19. Wentling, T.L., Waight, C., Gallaher, J., La Fleur, J., Wan, C., Kanfer, A.: E-Learning - A Review of Literature (2000), http://learning.ncsa.uiuc.edu/papers/elearnlit.pdf (retrieved March 31, 2007)

Pedagogical Strategy Model in Adaptive Learning System Focusing on Learning Styles Hongxia Liu*, Wei Zhao, and Ming Liang School of Computer Science and Information Technology, Northeast Normal University, 130117 Changchun, China {liuhx001,zhaow577,liangm471}@nenu.edu.cn

Abstract. Recently many educational researchers have proved that learning style is an important factor to impact on learning effect. This paper did a deep analysis on the characteristic and application of learning style models, chose FSLSM as the learning style model in our system, then drew up a plan to diagnose and revise the learning style. Based on the differences of learners' learning style, we construct the adaptive pedagogical strategy model under the Semantic Web from sections such as the learning resources, learning paths, adaptive rules. Keywords: Semantic web, Adaptive learning system, Learning style, Pedagogical strategy.

1 Introduction Recently more and more researches have proved that many factors can influence on learning, including learning background, learning environment, learning style, etc. But most of the Course Management Systems(CMS) typically neglect the individual differences of learners. Adaptive learning system is one of the effective solutions to solve this problem, which fits the individual characteristics of learners make learning easier for them. Adaptive Learning System (ALS), provides personalized service for learners based on the discrepancy during the learning process, such as personalized learning resources, learning paths, learning strategy, etc.[1] This paper discussed how to construct the pedagogical strategy model in adaptive learning system. This module mainly involves the following problems: (1)Choosing a learning style model as the criterion of the ALS and designing an effective, quantitative test paper to test learning style. (2) Diagnose the learning styles and determining the learners’ learning styles according to which kinds of information in learning behaviors. (3)How to realize * This paper is funded by the Ministry of Education in Humanities and Social Sciences planning project:Research on digital learning services model oriented to personal lifelong learning(08JA880012).This paper is also funded by technology development plan program in Jilin Province:Adaptive learning system under Semantic Web(20070521). X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 156–164, 2010. © Springer-Verlag Berlin Heidelberg 2010

Pedagogical Strategy Model in ALS Focusing on Learning Styles

157

adaptivity, exactly generating adaptive pedagogical strategies based on learning style. This paper explained these three questions in detail.

2 Learning Style Model First, we should choose a proper learning style model, and design an effective, quantitative paper to test learning style. Currently there are many types of learning style models from different angles. Coffield et al. (2004)[2]classified learning style models into 5 families from the angle of design. The first family classified learning styles and preferences into four modalities: visual, auditory, kinaesthetic and tactile. The second family deals with the idea that learning styles reflect deep level of the cognitive structure. In the third family, learning styles are seen as one component of a relatively stable personality type. The fourth family considers learning style as flexibly stable learning preferences. The last family moves on from learning style to learning approaches, strategies, orientations and conceptions of learning. These models are widely used in educational hypermedia systems. At the beginning of this study, we did a survey on learning style model of the mainstream Adaptive Educational Hypermedia Systems (AEHS), focusing on the dimensions of learning style models and adaptive strategies they have chosen. Table 1 lists the applications of learning style models in AEHS. Table 1. Learning style models in AEHS

Learning Style Model Felder & Silverman

Honey & Mumford Witkin Dunn & Dunn

Adaptive Educational Hypermedia Systems (Dimensions) CS383(sensing/intuitive, visual/verbal, sequential/global) Tangow (sensing/ intuitive, sequential/global) LSAS(sequential/global) PHP Programming course (four dimensions) INSPIRE (Activist/Theorist/Pragmatist/Reflector) AES-CS field dependence/field independence iWeaver (five perceptual, four psychological preferences)

（

）

We choose FSLSM(Felder & Sliverman Learning Style Model). First because of its reliability, till now, FSLSM has been confirmed by the author and many experts(Zywno, 2003; Felder & Spurlin, 2005),[3] also applied in various adaptive hypermedia education systems successfully, such as CS383, Tangow, PHP Programming course. These cases show that the model has good effectiveness and reliability. Secondly, FSLSM provides a quantitative Index of Learning Styles(ILS), a calculation model of four dimensions and these results are easy to understand and operate.[4] Furthermore FSLSM pays more attention to the media type, can give strong guidance to the pedagogical strategy based on learning style. Based on the these advantages, we choose FSLSM as learning style model of our adaptive learning system and use ILS to diagnose learning style of the learners. In our study, we did a survey on the applicability of ILS from two aspects: (1) FSLSM is suitable to traditional education, whether the preference through the test of ILS is equally applicable in online course. (2) Whether there is a strong correlation between

158

H. Liu, W. Zhao, and M. Liang

the results from ILS and the learners' online behavior patterns. According to the survey, we revised the adaptive pedagogical strategies, and formulated the behavior model we should monitor. This paper aims to construct the adaptive pedagogical strategy model, so this survey will not be illustrated in detail.

3 Learning Styles Diagnosis Once the model has been determined, the most urgent problem is how to determine learners’ learning styles more accurately. Normally, learning style diagnosis including two ways: explicit diagnosis and implicit revision. Explicit diagnosis mainly means guiding learners to fill in the questionnaires and learning styles revision mainly carry out through monitoring in the learning process. In adaptive learning system, we adopt a combined way to diagnose the learning style. Before entering the knowledge learning, we suggest learners take part in learning style test. If they disagree with the result of the test, the ALS provides an interface of “learning style switch”, then learners can adjust their learning style at any time in the learning process. At the same time, we draw up a plan for learning style revision based on the learning behaviors. 3.1 Explicit Diagnosis of Learning Style At this stage, we choose ILS to diagnose the learners’ learning styles. ILS divided learning style into 4 dimensions and 8 types from information processing, perception, input, understanding. The four dimensions are independent to each other: active/reflective, sensing/intuitive, visual/verbal, sequential/global. 3.1.1 Ex-test of Learning Styles Ex-test of learning style will initialize partial field of learning style module. ILS has strong operability. The ALS gets the learning style tendency and degrees after learners have filled in the ILS(learners’ learning style is described by their preference on each of the four dimensions, measured on values between +11 to -11, in steps of +/-2 ), then the ALS can deduce learning style of learners. 3.1.2 Learning Styles Switch Single questionnaire usually have certain errors, moreover some of the learners’ learning style tendency is not obvious. In ALS, the result of ILS is not the only criterion, but gives learners full freedom of choice. The system provides the function of “learning style switch”, so that learners can adjust their learning style according to their requirements. 3.2 Learning Styles Revision We test learning style for learners through ILS, and provide an interface of “learning style switch”, but both of them are subjective. Moreover, learning style of learners are not unchangeable and single questionnaire usually have certain errors, so it’s necessary to revise the learning style model from data mining during the learning process.

Pedagogical Strategy Model in ALS Focusing on Learning Styles

159

We monitor the actions of learners and modify learning styles through their behaviors. At this stage, we need to determine which kinds of behavior information can revise learners’ learning style. Table 2 lists behavior patterns relevant to each dimension of FSLSM. Table 2. Relevant behavior patterns for each learning style dimension of FSLSM

active/reflective forum_stay forum_post forum_visit exercise_visit example_visit

sensing/intuitive

visual/verbal

sequential/global

concrete content_visit concrete content_stay abstract content_visit abstract content_stay example_stay

text_visit text_stay video_visit video_stay image_visit image_stay

navigation_stay navigation_visit outline_stay outline_visit conclusion _stay conclusion_visit

3.2.1 Active/Reflective Dimension Active learners often take action without thinking deeply, express their ideas actively, and grasp information from communication or coordination. Reflective learners prefer to think quietly, express ideas after they had a comprehensive and elaborate analysis, tend to learning independently. In this dimension, we revise the learning styles model mainly through the participation in the forum, the frequency of browsing exercises and examples. 3.2.2 Sensing/Intuitive Dimension Sensing learners are glad to learn concrete material, pay attention to the details and be careful with their work. Moreover, sensing learners start a knowledge object more often than intuitive learners. Whereas intuitive learners prefer abstract material, like to innovate, discover possibilities and relationships. [5] In this dimension, we revise the learning styles tendency mainly through the abstract degree of materials they visited and time they spent on examples. 3.2.3 Visual/Verbal Dimension Visual learners are more sensitive to the images and like to acquire knowledge from pictures, charts, films, etc. Verbal learners have better understanding on information such as text or oral expression. In this dimension, we revise the learners’ learning styles mainly through recording the browsing situation of different media types. 3.2.4 Sequential/Global Dimension The main characteristic of sequential learners is that they learn in a linear way, going through the materials step by step. Global learners tend to prefer grasping the framework first and afterwards entering to the deep thinking. In this dimension, we revise the learners’ learning styles mainly through the browsing situation of navigation, outline and conclusions.

160

H. Liu, W. Zhao, and M. Liang

4 Generation of Adaptive Pedagogical Strategies There are both strengths and weakness in learning for each student's learning style. The basic purpose of education is not only to give full play to its advantages, but also try to make up its disadvantages in learning styles and learning tendencies. The shortterm goal of adaptive learning system is to improve the learning efficiency, and the long-term goal is to help learners adapt to all types of resources. In this study, we try to find a learning service model for learners with different learning styles. What we mainly consider is how to provide learning resources, learning activities and learning paths to suit the learner's personalization features,[6] that is, to construct a reasonable model of pedagogical strategies. 4.1 Presentation of Adaptive Content Traditional learning systems are often lack of effective description of their own learning resources, limiting the reasonable presentation of the learning content. And it’s difficult to solve this problem in the condition of traditional database modeling. In Semantic Web environment, we can easily add the attributes for the resources with Table 3. The semantic metadata and its impact associated with learning styles

Attributes of Metadata

Instructions

Dimensions Mainly Affected

MediaType

Including types of text, pictures, video, audio, animation, etc.

Abstractness

The presentation of knowledge point is concrete or abstract ˄ For example, the abstract level of a concept is often higher than an instance˅.

Sensing/Intuitive Active/Reflective

Exercises

The exercises related with knowledge points.

Active/Reflective Sequential/Globa l

Examples

Interpretation or application of knowledge points

Sensing/Intuitive Active/Reflective

Visual/Verbal

Knowledge Presentation Presenting pictures, video and animation mainly to the visual learners, and text, audio to the verbal learners. Presenting abstract content to the sensing and reflective learners, concrete content to the intuitive and active learners. Suggesting the active and sequential learners doing more exercises. Suggesting the sensing and reflective learners studying mainly through examples

Pedagogical Strategy Model in ALS Focusing on Learning Styles

161

semantic metadata by using the ontology modeling, [7] which provides a good solution to this problem. Considering a variety of factors involved in learning styles, we define some metadata information such as knowledge types, media types, abstractness, etc. Table 3 lists some of the semantic metadata associated with learning styles. As can be seen from Table 3, the system provides the corresponding presentation of knowledge for the learners according to their learning styles. In order to make the learners not limited to the content directly shown to them, we provide an interface to other high-quality resources for the learners so that they can effectively use the additional resources related to the knowledge point. 4.2 Presentation of Adaptive Navigation Rational learning paths has an important impact on the cognitive processes for the learners. For example, the global learners are likely to acquire the overall framework of knowledge and then refine it gradually, which is opposite to the sequential learners. ALS provides personalized learning paths to learners with different learning styles by using adaptive navigation. In Semantic Web environment, we set attributes such as KnowledgeType, Primary, Further, Difficulty, Related and hasParent/hasChild for each knowledge point, so that learning content can be formed into a unified, standardized, clear semantic structure. It achieves effective reasoning between knowledge points and provides possibility for the realization of adaptive navigation. The navigation in Adaptive Learning System can be divided into global navigation and local navigation. We show the global navigation mainly by tree structure of domain knowledge. The tree structure can show complete knowledge of the course system, and the current state of knowledge mastery to the learners through learning state markers. Through the global navigation, all types of learners can identify the situation of their current learning content in the whole knowledge system and the mastery of the curriculum, thus avoiding the information disorientation and ambiguous situation. Local navigation is the key of adaptive navigation. Learners of different learning styles get personalized learning paths right through the local navigation. The ALS provides navigation map, showing learning paths for learners. The learning steps are arranged from top to bottom in order of priority, and the learners can either learn according to the proposed steps, or make their own arrangements. The ALS designs learning paths in detail for learners of each learning type. Take the arrangement of learning paths with active/reflective dimension for example, the system presents the overview of knowledge for learners at first, so that the learners begin to study the learning objects in depth after they have grasped the overall knowledge. And then they have a summative review, followed by judgment of the system on active/reflective dimension. When the learners are judged as active type, the system presents exercises first and examples as follow, because of learners of this type liking to learn from doing exercises. For the reflective learners, the order of presentation is on the contrary. Then, learners can access the forum to discuss and exchange ideas with others, and at last they do tests of evaluation and make up the shortage in the study. (Fig 1). Local navigation provides concept map of knowledge for learners so that they can easily access the relevant knowledge, primary knowledge and the further knowledge etc, helping them form reasonable schema.

162

H. Liu, W. Zhao, and M. Liang

Fig. 1. Arrangement of the learning paths

4.3 Adaptive Sort of the Resources We provide learners with personalized learning plans through the presentation of adaptive knowledge content and learning paths. And we need to ensure that the learners with a certain style could learn high-quality learning resources in priority after the manner of resource presentation having been determined. So it is necessary to make effective evaluation and sorting for the learning resources. As a necessary complement for personalized learning plans, adaptive sort for learning resources provides optimization of resources for learners with a certain style. In the adaptive learning system, we have adopted the intelligent mining techniques under the concept of web2.0, and one of the most typical applications is Digger. The core idea of Digger is to launch people to excavate and the order of resources presentation is determined by the users’ behaviors. We transplant the Digger into the ALS so that the traditional top-down resource presentation system leaded by a handful of experts becomes bottom-up adaptive resource system whose main control factors are the vast number of learners. As the functions and services of the system, Digger is realized mainly through the evaluation of learning resources given by the learners. The ALS set the corresponding weights for the evaluation of academic experts and learners.

Pedagogical Strategy Model in ALS Focusing on Learning Styles

163

Due to the lack of adequate users, the academic experts play a major role in the evaluation of resources at the initial stage of the system, which solves the "cold start" problem in a certain degree so that learners can also browse the high-quality resources in the beginning. With the increase of the number of learners, the system adjusts the weight distribution dynamically, and the evaluation of learners will replace the experts gradually, becoming the main factor of resource evaluation and sorting. The evaluation of resources will be more and more objective and stable. 4.4 Adaptive Rules The adaptive rules of the ALS provide common ways and methods to present knowledge for learners with different learning styles. The same set of adaptation rules may result in a different presentation depending on the adaptive system with different user models. [8] The adaptive learning system under the Semantic Web environment is composed of nodes (pages) and links. Except for the content, each page contains some links pointing to other pages. [9] In the traditional learning systems, the relationship between the links and the content of knowledge is fixed, while the dynamic relationship between the two parts is the main foundation to achieve the adaptive rules in the ALS. 4.4.1 Transparent Presentation of Information The page presents the users contents of the suitable style directly after the system having judged the learners’ type of learning style. In other words, learners of different styles see different page content for the same knowledge point, while they cannot see the recommendation process of the system. 4.4.2 Selective Hiding of Information The system shows the most suitable knowledge content to the learners according to their characteristics while some content and links are set to be hidden. For example, we can offer corresponding exercises for each knowledge point in the same page and set them to be hidden. After taking the test of the chapter, the learners may find that some knowledge points being not mastered well. And when they re-learn the knowledge points, the corresponding exercises will be shown so that the learners could have self-tests conveniently. 4.4.3 Providing Explanation Variant The ALS keeps two or more variants of the same page with different presentations of the same content. When presenting a page, the system selects the page variant according to the users’ learning types. [10] For example, we can offer the learners different textual expression for the same concept, and we can present description of a flow in different methods such as flow chart and text, etc. 4.4.4 Providing Learning State Markers Learners need to know their learning state in the process of learning such as the mastery of the curriculum, so the system marks the learning progress for the learners. For example, we can mark the mastered knowledge in the knowledge tree structure, the learning knowledge and the proposed knowledge by using different color markers.

164

H. Liu, W. Zhao, and M. Liang

4.4.5 Providing Links of Relevant Knowledge The learners do not master a knowledge point well sometimes because of their failure to master the relevant knowledge. For instance, they may have not learned the primary knowledge thoroughly, or they may have confusion on the similar concepts. Based on the building of domain knowledge structure, the system provides links of related knowledge for the learners so that they can easily jump to the points of relevant knowledge.

5 Conclusion Semantic Web is the intelligent networks of the next generation, and adaptive learning in the Semantic Web environment will also be an important trend with the development of intelligent learning. It is the foundation of the system function and application to construct the adaptive pedagogical strategies model for learners with different learning styles. This study is a bridge between the user model and the domain model in the ALS. And they are the three core components of the ALS. The related study such as the adaptive evaluation based on learning effects will be reflected in the subsequent articles.

References 1. Boping, H., Wei, Z., Yandong, Y.: Contrastive Analysis on Adaptive Learning System Model. J. China Educational Technology (8) (2009) 2. Graf, S.: Adaptivity in Learning Management Systems Focussing on Learning Styles. D. University of Vienna (2007) 3. Popescu, E.: Dynamic adaptive hypermedia systems for e-learning. D. University of Craiova (2008) 4. Graf, S., Viola, S.R., Leo, T., Kinshuk: In-Depth Analysis of the Felder-Silverman Learning Style Dimensions. J. Journal of Research on Technology in Education 40(1) (2007) 5. Baishuang, Q., Wei, Z., Xiuqin, L.: The Research of User Model in Adaptive Learning System Based-on Semantic Web. J. Open Education Research (4) (2008) 6. Wu, H., De Kort, E., De Bra, P.: Design Issues for General - Purpose Adaptive Hypermedia Systems. In: Conference on Hypertext and Hypermedia, pp. 141–150 (2001) 7. Henze, N., Dolog, P., Nejdl, W.: Reasoning and Ontologies for Personalized E-Learning in the Semantic Web. J. Educational Technology & Society 7(4), 82–97 (2004) 8. Berlanga, A.J., García-Peñalvo, F.J.: Learning Design in Adaptive Educational Hypermedia Systems. J. Journal of Universal Computer Science 14(22), 3627–3647 (2008) 9. Pinde, C., Kedong, L.: A Review of Adaptive Hypermedia System: Model, Method and Technology. J. Modern Educational Technology (1) (2002) 10. Brusilovsky, P.: Methods and techniques of adaptive hypermedia. User Modeling and User Adapted Interaction J. 6(2-3), 87–129 (1996)

Transferring Design Knowledge: Challenges and Opportunities Jun Hu, Wei Chen, Christoph Bartneck, and Matthias Rauterberg Designed Intelligence Group, Department of Industrial Design Eindhoven University of Technology Den Dolech 2, 5612AZ, Eindhoven, The Netherlands

Abstract. Design becomes more and more the art of bringing together expertise and experts from different domains in creating future products. Synthetical knowledge and hands-on skills in design, especially in industrial design, is often implicit, hardly captured and modeled for remote education. The need of transferring implicit design knowledge using computer mediated learning tools, provides not only technical challenges, but also many research opportunities. In this article the literature about training transfer and implicit design knowledge transfer is reviewed. A scenario of using such learning tools for learning and teaching physical modeling in industrial design is presented, followed by a discussion about the challenges and opportunities in developing such a system.

1

Introduction

Design becomes more and more the art of bringing together expertise and experts from different domains in creating future products. For example, knowledge from the fields of engineering, ergonomics and psychology are needed for designing future aircraft seat [20]; while designing the next generation of health monitoring systems requires the integration of expertise from medical science, electrical engineering and user research [7,6]. This multi-disciplinary character requires changes of the design environments, methods and tools in all phases of the development cycle. Besides the analysis based approach, effective and adequate design synthesis is required. In this, synthesis implies composing design elements and approaches, so as to form a whole, as well as combining diverse conceptions into a coherent whole. Design synthesis therefore relies on e.g. adequate process descriptions and methods, on structured means for information and knowledge management, on effective (analysis and simulation) tools, on stimulating (virtual reality) environments and on suitable methods to govern the entire process. But the decisive factor is the effective and far-reaching facilitation of the interaction between designers, domain experts and users. Most of the computer-aided learning systems are focused on the analytical knowledge and skills. Synthetical knowledge and hands-on skills in design, especially in industrial design, are essential for designers, which requires the education and the learning process to adapt to this need [14,22,13,1]. These knowledge and skills are often implicit, hardly captured and modeled for remote education. For example, as key steps in industrial design, physical models are created to X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 165–172, 2010. c Springer-Verlag Berlin Heidelberg 2010 

166

J. Hu et al.

visualize the ideas and concepts, to engage the end users and the domain experts for their feedback (for example in [18]). Skills in creating these models are essential, but difficult to be learned from text books. They are learned in practise by close observation and through direct hand-to-hand training. Such a process is expensive and not scalable with the number of students and their geographic distribution. From the pedagogical point of view, modeling is very complex as it involves a number of dimensions. Considering the Bloom’s taxonomy [4], it involves all the three domains: cognitive, affective and psychomotor. In this article we first review the literature about training transfer, especially about implicit design knowledge transfer. We then describe a scenario in which implicit design knowledge would be transferred over a computer mediated system. From this scenario requirements for such a system are identified, followed by a discussion about the challenges and opportunities in developing such a system.

2

Training Transfer and Virtual Training

Because organizations consider human capital one of the most salient organizational assets in establishing and maintaining a competitive advantage, many are investing considerable resources to support employee and organizational development activities such as training [3,19]. Globalization, technological advancements, and talent wars in recruiting and retaining high performers are among the other major reasons that organizations seek to leverage training outcomes to foster workplace performance improvement, facilitate development of individual and organizational effectiveness, and establish and maintain market share within the rapidly changing business environment [5,16]. To identify how training interventions influence performance outcomes in organizations, many researchers have conducted training transfer studies, which have revealed numerous findings about the effect of such transfer factors on employee and organizational performance outcomes [12]. Traditional training transfer studies have assessed the separate influence of trainee characteristics, training design, and work climate variables on training transfer in attempting to validate the influence of each of these independent variables on training transfer [2,8,12]; few have used integrated approaches and examined the empirical assessment of cross-relationships and influence of those diverse variables in trainees’ characteristics, work and job experiences, tool support, and organizational climate on learning and transfer outcomes [21]. The need to identify the mediating mechanisms to link contextual features influencing training transfer has been a pressing research interest among several researchers [15]. For practitioners, identifying good research findings about the integrative and interactive features among different variables in trainees, training design, tool support, and organizational climate can be valuable in improving the effectiveness of delivered training. Among research in education tools done to date, several approaches such as CAI, ICAI, Micro-world, ITS, ILE, and CSCL have been proposed and many systems have been built within each approach. Innovative technologies such as

Transferring Design Knowledge: Challenges and Opportunities

167

hyper-media, virtual reality, internet, WWW have significantly affected the research community in general. Recent advances are offering an increasing number of innovative and promising learning environments including three-dimensional and two-dimensional virtual worlds as well as computer simulations. These environments differ a lot as to both their technological sophistication and to the types of skills taught, varying for example from immersive 3D environments with haptic feedback of high-fidelity to simulations of complex relational situations, for the learning of “soft design skills” of growing strategic interest to enterprises such as leadership, customer service, coaching, selling etc. The learning potential of virtual training relies on the possibility for trainees to make a number of significant first-person experiences and to fail in a safe and protected environment. In order to be effective, the training experience should seem real and engaging to trainees, as ‘if they are in there’ they should feel (psychomotor, emotionally and cognitively) present in the design situation. It is therefore important to investigate the relationships existing among the factors that are crucial to the transfer of implicit design knowledge in virtual training environments. There is a need to develop a sophisticated tool support in virtual learning environments, trying to define on the one hand the key factors conveying it in training contexts and on the other hand how the tool support contributes to enhance learning efficacy and to support transfer of design knowledge and skills, which is depicted in the following scenario.

3

Scenario

This scenario is based on the teaching and learning practice in cardboard modeling. Typical activities for a student to master such a technique are [10]: acquire basic skills in constructing a set of basic forms and mechanisms providing action possibilities; learn to capture the essence of the complex form of existing products, plan the mock-up and construct it accordingly; learn to use modeling for exploring the solution space of a design problem. Joe is an expert in modeling. He teaches cardboard modeling at a design school in Eindhoven. He usually provides the course every semester for a small group of maximum 10 students because of the hands-on nature of the course, but now he has to provide the course for up to 100 students and he is also invited to give the course for another design school in Hong Kong. Figure 1 and 2 are examples of cardboard models made for exploring rich interaction possibilities. Joe decides to use the Virtual Trainer (VT) toolkit to face this new didactic challenge. He prepares his course by first making several cardboard models, from simple ones to complex ones. The VT tools record his actions and behaviors in the process of the modeling. These actions are recorded along a timeline in a multimedia digital format. The format integrates an audiovisual stream and semantically annotated data that describe the teacher activity and important properties (i.e. the status) of the physical model as they evolving during the artefact development time. The multimedia material is prepared automatically by recording, analyzing, filtering and composing data.

168

J. Hu et al.

Fig. 1. Cardboard model of a redesigned CD player [11]

Fig. 2. Cardboard models of cameras designed for rich interaction [9]

When finished with the models, Joe reviews the material and edits some additional personal instruction. Then he packs all the prepared digital material and make it available to his students in Eindhoven and Hong Kong. He asks all his students to follow the online instructions to practise by themselves. Jackie is one of Joe’s students in Hong Kong. Following the online instructions, his assignment today is to make a cardboard model of a camera (Figure 2). He prepares his digitally enhanced tools (the VT toolkit) by loading the digital materials received from Joe. The wearable tools allow Jack to see the cardboards augmented with the cut lines indicating steps, tools and actions. When cutting the cardboards, the tool gives immediate feedback on his actions, on whether a right tool is used, whether it is cut in a right direction, with the right force etc. Suggestions are provided if corrections/improvements are needed. The intermediate and finalresults are compared to the 3D model and analyzed immediately; feedback and explanations are given if necessary. Multimodal feedback is provided in textual, visual and haptic ways, in order to involve the user in the most suited way. In this scenario, a learning appliance that supports teaching and learning in 3D industrial design modeling plays the central role. The appliance consists of a tool that records and annotates a teacher’s actions and more importantly, his behaviors, in making a cardboard model, then instructs the students to create the model in a similar manner.

Transferring Design Knowledge: Challenges and Opportunities

169

Instructions and feedback The appliance consists also of a complementary set of tools, for example wearable devices to be used by students, which allow them to get realtime multimodal feedback about the work they are doing. The feedback Experienced designers is based on the comparison between Students their model and the way the actions they are taking with the model and Observations and questions the actions of the teacher (see Figure Fig. 3. Traditional learning model 3 and 4 where the traditional and proposed training models models are depicted). The tools also support communication and remote presence, fostering cooperation among students and strengthening interaction (remote and local) with the teacher. One must notice that the Model in Figure 4 is crucial for capturing the actions and behaviors of the teacher, being transferred over distance, and being utilized for training the students in either real time, or for a later time.

4

Challenges and Opportunities

From a technological point of view, the VT learning model involves a number of challenges in order to meet the ambitious objectives. These challenges can be tackled with state of the art technologies provided that the system design is well directed. Enabling technologies include: – Sensors to detect and characterize the users’ (teachers and students) actions and detailed features of the models as they are being prepared. These include for example cameras, laser scanners and wearable sensors. – Sensor fusion to coordinate data from various sources and automatic generation of 3D models from the sensors. – Accurate modeling of the teaching and learning activities (considering the three Bloom domains: cognitive, affective and psychomotor) and of the produced/under production artefacts Instructions and feedback

Capturing system

Model

Model

Virtual Trainer

Experienced designers Students Observations and questions

Fig. 4. Proposed learning model

170

J. Hu et al.

– Automatic annotation of the human design activities in the digital multimedia stream (according to the user cognitive modeling), for example using ontological engineering approach to externalize the features of the empiricallyformed wisdom or skills of a designer digitally. – Artificial Intelligence and Machine Learning algorithms in order to train the tools so that they are able to evaluate student activities and their products as compared to teacher’s references (also providing feedback for correction and improvement), based on the user cognitive model. – Human-Computer Interaction in order to make the tools as useful and usable (effective, efficient, pleasant to use) as possible. User interaction modalities also include augmented reality wearable devices and haptic feedback. From the pedagogical and psychological points of view, such a system can enhance learning processes and skills acquisition, which brings many opportunities to the education and the industry: – Since the tool can overcome the single tutor-trainee interaction and can be used by many students, whenever and wherever they want, it gives a great training opportunity due to the repeatability of the task; this allows learning to be faster and more effective, since acquiring skills about such complex procedural tasks is grounded on students’ management of their learning time and dynamics. – The system is equipped with teaching support systems and hints implemented according to the guidelines provided by subject matter experts in order to balance reflexive and experiential thinking in doing the task. – It improves continuous feedback between the student and the virtual tutor, so that each step in learning acquisition can be checked and be part of a whole coherent mental model. – The interface is designed according to cognitive ergonomics principles of learning transfer, usability and ecological design, to allow students to live a rich and stimulating experience, which will set the stage for basic skills acquisition, and as a further step, a proper ground for creativity development. – It helps to establish valuable archive/database for interactive teaching and learning, for example, providing model samples and common error examples. The database can be used for evaluating the effectiveness of the teaching and learning process. It is interesting to utilize an ontological engineering approach analyzing designers’ modeling explanation style based on the observable psychomotor design activities. In an activity based ontological engineering approach, defining knowledge as ontology is effective in extracting its essential qualities. it is then interesting to investigate the effectiveness of this new ontological engineering approach (see also [17]), resulting in an ontological framework of designers’ explanations: “clarification of the essence of the design style”, “discovery of problems in explanations”, and “analyzing difficulties in acquiring design style for trainees”. The system also provides the opportunity to investigate the relationships existing among the factors that are crucial to the transfer of implicit design knowledge

Transferring Design Knowledge: Challenges and Opportunities

171

in virtual training environments. In reviewing the training transfer literature, we found that some transfer studies had occurred in international settings in countries in Africa, Europe, and Asia. However, in our review, we seldom found evidence suggesting that studies have examined the empirical assessment of cross-relationships and the influence of those variables. Therefore, developing such a system is unique in that it identifies the cross-relationships and effect of the multi-domain transfer variables on training transfer, and addresses training transfer issues from an international context. The results will prove valuable in expanding the scope of existing research studies of training transfer at the international level based on advanced tool support.

5

Conclusion

For transferring implicit design knowledge using computer mediated learning tools, such as the Virtual Trainer system described in this article, requirements are identified and challenges and opportunities are discussed. Among other challenges, how to model of the implicit design knowledge is essential. It is also important to identify the effect of transfer design knowledge according to the relative importance of this knowledge in their influence on trainees’ learning, perceived learning applicability, and perceived learning application. Verification of transfer results from post-transfer data such as design performance data will be an important consideration of validation studies. Through these studies, it is possible to reveal meaningful findings in cross-construct settings if other variables in motivation, organizational reward systems, and work ethics from different organizational and cultural settings are used. It is also possible to examine the multi- or uni-dimensional characteristics of design trainer variables as they influence the training transfer processes, and identify when and why certain design trainer variables become converged or distinct under what transfer situation. The need of the virtual training tools for transferring implicit design knowledge provides not only technical challenges, but also many research opportunities.

References 1. Alers, S., Hu, J.: Admoveo: A robotic platform for teaching creative programming to designers. In: Chang, M., Kuo, R., Kinshuk, Chen, G.-D., Hirose, M. (eds.) Learning by Playing. Game-based Education System Design and Development. LNCS, vol. 5670, pp. 410–421. Springer, Heidelberg (2009) 2. Baldwin, T.T., Ford, J.K.: Transfer of training: A review and directions for future research. The training and development sourcebook, 180 (1994) 3. Becker, B.E., Huselid, M.A., Ulrich, D.: The HR scorecard: Linking people, strategy, and performance. Harvard Business School Pr., Boston (2001) 4. Bloom, B.S.: Taxonomy of educational objectives. Handbook I: Cognitive Domain (1956) 5. Branham, L.: The 7 hidden reasons employees leave. American Management, New York (2005)

172

J. Hu et al.

6. Chen, W., Bouwstra, S., Oetomo, S.B., Feijs, L.M.G.: Intelligent Design for Neonatal Monitoring with Wearable Sensors, pp. 386–410. INTECH (2010) 7. Chen, W., Hu, J., Bouwstra, S., Oetomo, S.B., Feijs, L.M.G.: Sensor integration for parinotology. Accepted by International Journal of Sensor Networks IJSNet Special Issue on Recent Advances in Sensor Integration (2010) 8. Ford, J.K., Weissbein, D.A.: Transfer of training: An updated review and analysis. Performance Improvement Quarterly 10, 22–41 (1997) 9. Frens, J.: A rich user interface for a digital camera. Personal and Ubiquitous Computing 10(2), 177–180 (2006) 10. Frens, J.: Dg101 - cardboard modelling, assignment description (2009), https://venus.tue.nl/owinfo-cgi/owi_0695.opl?vakcode=dg101 11. Hendriks, B., Hu, J.: Redesigning a cd player for intuitive rich interaction. In: 12th International Conference on Human-Computer Interaction, CD Proceedings, pp. 1607–1611. Springer, Heidelberg (2007) 12. Holton, E., Bates, R., Ruona, W., Leimbach, M.: Development and validation of a generalized learning transfer climate questionnaire: Final report. In: Proceedings of the 1998 Academy of Human Resource Development Annual Conference, Academy of Human Resource Development, Chicago, IL, pp. 482–489 (1998) 13. Hu, J., Alers, S.: Admoveo: An educational robotic platform for learning behavior programming. In: DeSForM 2009: Design and Semantics of Form and Movement, Taipei, Taiwan, pp. 218–219 (2009) 14. Hu, J., Ross, P., Feijs, L., Qian, Y.: Uml in action: Integrating formal methods in industrial design education. In: Hui, K.-c., Pan, Z., Chung, R.C.-k., Wang, C.C.L., Jin, X., G¨ obel, S., Li, E.C.-L. (eds.) EDUTAINMENT 2007. LNCS, vol. 4469, pp. 489–498. Springer, Heidelberg (2007) 15. Kozlowski, S.W.J., Farr, J.L.: An integrative model of updating and performance. Human Performance 1(1), 5–29 (1988) 16. Michaels, E., Handfield-Jones, H., Axelrod, B.: The war for talent. Harvard Business School Pr., Boston (2001) 17. Mizoguchi, R., Bourdeau, J.: Using ontological engineering to overcome common ai-ed problems. International Journal of Artificial Intelligence in Education 11(2), 107–121 (2000) 18. van de Mortel, D., Hu, J.: Apartgame: a multiuser tabletop game platform for intensive public use. In: Tangible Play Workshop, Intelligent User Interfaces Conference, Honolulu, Hawaii, USA, pp. 49–52 (2007) 19. Pfeffer, J.: Competitive advantage through people: Unleashing the power of the work force. Harvard Business School Pr., Boston (1994) 20. Tan, C., Chen, W., Kimman, F., Rauterberg, M.: Sleeping posture analysis of economy class aircraft seat. In: Accepted to World Congress on Engineering 2009, London, U.K (2009) 21. Tracey, J.B., Hinkin, T.R., Tannenbaum, S., Mathieu, J.E.: The influence of individual characteristics and the work environment on varying levels of training outcomes. Human Resource Development Quarterly 12(1), 5 (2001) 22. van der Vlist, B., van de Westelaken, R., Bartneck, C., Hu, J., Ahn, R., Barakova, E., Delbressine, F., Feijs, L.: Teaching machine learning to design students. In: Pan, Z., Zhang, X., El Rhalibi, A., Woo, W., Li, Y. (eds.) Edutainment 2008. LNCS, vol. 5093, pp. 206–217. Springer, Heidelberg (2008)

The Content Balancing Method for Item Selection in CAT Peng Lu1,3, Dongdai Zhou1,2,3,4, Xiao Cong1,3, Wei Wang1,3, and Da Xu1,3 1

Ideal Institute of Information and Technology, Northeast Normal University, China, 130024 2 School of Software, Northeast Normal University, China, 130024 3 Engineering & Research Center of E-learning, China, 130024 4 E-learning Laboratory of Jilin Province, Changchun, Jilin, 130024 {lup595,ddZhou,Congx805,wangw577,xud169}@nenu.edu.cn

Abstract. Compared with traditional testing, Computerized Adaptive Testing owes incomparable advantages. Such as flexibility, reduce the test length and measurement accuracy. There are some components in CAT, the most one is the item selection algorithm. To perform adaptive test, the most frequently adopted method is based on the maximum information (MI) of items to select the examination questions, with the view to draw the most accurate estimation for tester’s capacity. There exists, however, flaws of unbalanced item-exposure as well as unequalled usage of item pool in this method. In this paper, we propose a new item selection algorithm CBIS to solve those problems, and then compare our method with MI method by an experiments. The experiment results are promising. Keywords: Adaptive Testing; IRT; Content Balancing; Exposure Rate.

1 Introduction Learning diagnosis has always been a very important step in the learning process [1]. In the adaptive learning system, the tests are required to measure students’ learning outcomes as much as possible, and meanwhile with high efficiency, diagnosing students’ real capacity with the minimum test items. And the diagnostic results, under various conditions and within different period of time, must be in accordance with one another. In the CTT, ever students take the same test, with fixed test length and time. Therefore, testing is not flexible enough. The difficulty of question targeted at the average level of students, the results are not immediately visible. Computerized Adaptive Testing (CAT) is a test administered by a computer, where the selection of the next question to ask and the decision to stop the test are performed dynamically based on a student profile which is created and updated during the interaction with the system [1]. In 1970, Lord defined the theoretical structure of Machine Adaptive Testing [2], which could give tests not only for students group but also every learner. In 1990, Wainer & Mislevy made a thorough discussion on Machine Adaptive Testing from the philosophic angle [3], “The basic notion of an adaptive test is to mimic automatically what a wise examiner would do.” Currently, Adaptive Testing has already been widely X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 173–184, 2010. © Springer-Verlag Berlin Heidelberg 2010

174

P. Lu et al.

used in many areas, such as educational assessment, psychological test, talents selection, and medical diagnosis, etc. [4]. In 1950s, there gradually developed the modern test theory- Item Response Theory (IRT) [5,6], and the testing model based on this theory was regarded as IRT Model, which defined the relations between the tester’s response to items and his latent trait. Choosing items that can match tester’s capacity according to the amount of item information provided until it meets the predetermined accuracy requirements. The three-parameter logistic (3PL) model is the most widely used one, and its formula is as follows:

P(θ) = c + (1 − c)

1 1+ e

− Da(θ - b)

.

(1)

where D=1.702, a constant; θ, the estimate value of tester’s capacity; a, the item discrimination; b, the item difficulty coefficients; c, the item pseudo-guessing parameter; P(θ), the probability of a person with θ capacity to response correctly, that is the probability of accurate reflection. In the adaptive testing, it is of little importance to select any item as the first one [7], for it will not exert any crucial influences on the final testing results. In general, the first item should be comparatively easy to give the testers a sense of accomplishment [8]. Compared with traditional test, the adaptive testing is characterized with the following aspects: unrestricted test time, different test length and items for different tester, and immediate visible test outcomes. Thus, CAT can be a more appropriate evaluation system, which can constantly calculate the tester’s current ability value based on his response, and then to instantly adjust item level according to these parameters, until the tester’s ability to be properly assessed. The main components of an adaptive testing include design of item bank, the item selection method, the input proficiency level and the termination criterion etc.. Among them, item selection algorithm is the key. In recent years, a widely used method is the maximum information method (MI), which selects the next item with the largest Fisher information evaluated at the current ability. This approach can provide the most efficient ability estimation. However, it has been noted that the MI approach can result in skewed item exposure distributions, content balancing and the security of item pool etc. The paper aims to provide an item selection algorithm, which called CBIS (Content Balancing Item Select); it can solve the problems existing in present computerized adaptive testing, such as unbalanced test contents, high item exposure rate, uneven item utilization in item pool, and so on. Giving due considerations to content balancing and item exposure rate issues throughout the process of item selection, would, on the one hand, pay close attention to testing security and accuracy; on the other hand, get computers to execute algorithm with high efficiency, to reduce time consumption. The work is structured as follows. In section 2 we introduce an overview of the Item selection approach in CAT. The section 3 is devoted to showing the item selection algorithm of CBIS. In Section 4 is the experiment result of CBIS and MI. Finally, the last Section is devoted to conclusions and future work.

The Content Balancing Method for Item Selection in CAT

175

2 Overview of the Item Selection Approach in CAT In the CAT, item selection algorithm is the most essential part. And the most widely adopted item selection algorithm currently is the method of the maximum information of items (MI) [9], as shown in the formula (2), which can select and determine the next test items according to the tester’s present capacity. 2

I i (θ' ) =

[ Pi ' (θ ' )] . Pi (θ)[1 − Pi (θ ' )]

(2)

Based on the tester’s present ability level θ, this method can real-time calculate the maximum information of every item in item pool to select the continue test item [9]. The advantage of this method is its high testing accuracy, but the MI-based item selection algorithm may easily causes the problem of unbalance item exposure rate [10]: those items with high discrimination also share rather high exposure rate, while those with low discrimination would usually not be selected as testing items [11]. For one thing, due to the high exposure rate of some items, the testers can collect some data related to those items in advance, which, in turn, will affect the item security and influence the final test results. For another, as it would cost tremendous manpower and material resources to develop and maintain item pool and items, then the utilization of item pool may directly influence the cost price [12]. In addition, content balancing is an important issue in CAT [13]. A large number of adaptive testing algorithms don’t take test contents into consideration, and so dose the item selection strategy. It doesn’t consider content domain in subjects either. But in the actual tests, teachers or testers often tend to concerned about the proficiency of a certain or some of the contents. As a result, if the content balancing problems cannot be well resolved, then the accuracy of the final test results would be lowered, and the test objectives couldn’t be fulfilled [14]. Besides, through out the process of using the item pool, the parameters for the items are certain, which is contradict with the whole testing theory. With the ever increasing exposure rate of the items, more and more people get to know the item knowledge, therefore, the item difficulty should be in a decreased state or few people can get the correct answer of the same item in the whole testing process. In this case, in order to receive the best test results, the item difficulty should be adjusted dynamically. To solve the content balancing problems in adaptive testing, a two-level content evaluation algorithm based on the test contents has been proposed, which had also been applied in the CBAT-2 system [13]. Compared with MI-based algorithm, this one has made some progress in content balancing. It, yet, only concerns about the weighting of the two-level contents, and cares nothing about the weighting of its deeper sub-contents. So the selection of the items is not sufficiently precise, thus affecting the test results. Furthermore, to solve the flexible content balancing restriction, a two-stage item selection strategy MPI algorithm has been proposed [15], which will cope with lower boundary of every part’s content in the first stage of item selection, and then deal with the higher boundary in the second stage. This method can handle content balancing problems quite well with rather high calculating efficiency. However, as a heuristic method, it cannot guarantee its excellence throughout the whole project selection process [4].

176

P. Lu et al.

The AS approach has been shown to be effective in both reducing the item overlap rate [16]. Compared with the MI method in the whole testing process, AS method, first of all, would sort item’s order according to the discrimination, and then select those with low discrimination to test in the early stage; as the tests carry on, it, then, would select items with high discrimination. The experiments results prove that the AS method has great effects on reducing item exposure rate and increasing item pool utilization rate [16,17], as well as maintaining the test accuracy. Yet, it does not take into account the relationships between item difficulty and discrimination, mistakenly equaling item difficulty to item discrimination, thereupon, influences test results in the low-capacity tests. In order to solve the problem mentioned above, the AS strategy has been modified and developed the a-stratified with b-blocking method (BAS) [18], in which items in item pool would be firstly sorted in ascending order according to their difficulty b, being divided into several layers, and items in every layer would then be sorted based on discrimination a. But both AS and BAS method do not consider the test content, so in the process of item selection, the calculation is complex and the content balancing and item pool utilization rate have, as well, been affected. According to the research from above we can be found that researchers have proposed various item selection methods， but these methods are all improve in a particular aspect of item selection process. While there has been some progress, but ignore other aspects, and are not fully resolved the problems in the item selection process, which are the content of imbalance, a high exposure rate of portion of the questions, a low utilization rate of the item bank and test security.

3 Content Balancing Based Study of Adaptive Testing Algorithm In order to solve the shortcomings have proposed in the selection process, in this paper we propose a new approach of item selection， which called CBIS (Content Balancing Item Select) selection algorithm. The core of the algorithm is based on the weights of Item in the content area. 1) Giving due considerations to content balancing and item exposure rate issues throughout the process of item selection, would, 2) on the one hand, pay close attention to testing security and accuracy; 3) on the other hand, get computers to execute algorithm with high efficiency, to reduce time consumption. To begin with, the organization of item pool is in strict accordance with the scientific division of the specific subjects in content, and then forms a content tree according to contents’ interrelations and their importance, to determine the weighting, in the hope of providing the content balancing basis for testers to select items in the testing process. And meanwhile, enter relevant items according to. In the testing process, a pre-test is needed at the very beginning for testers to ascertain their original relevant parameters, and then the concrete tests would be conducted, and the test outcomes would be reached in the end. 3.1 Hierarchical Organizational Structure of Subjects Domain models are composed of basic subjects, and each unit can be further split into sub-unit and specific learning contents. At last, the item pool is organized based on

The Content Balancing Method for Item Selection in CAT

177

domain models. The rectangles in the above figure1 represent the course units of a certain content domain, and the leaf nodes represent the specific learning contents; while between the unit and sub-units there is a paradigmatic relation. Each node has a unique parent node, and root node has no parent node. Each testing item is related to a Topic/Topics or Concept/Concepts; the same item may be connected with different Topics or Concepts. If an item is applied to test Topic i, then it can also be used to test its parental Topic, or even the whole course. As shown in the above figure 1: Q8 is used to test Topic21, and it can also test Topic2.

Fig. 1. Structure of the Content Areas and Items

Fig. 2. The system's view in a Topic test

178

P. Lu et al.

In the whole course structure, the weighing is employed to represent the sub-node’s shared proportions in parent nodes. As for the content weighting, content structure organization and item entering are determined by the domain experts. When being tested, the test content should be fixed firstly. By this way, for one thing, it is not necessary to calculate maximum information of each item in the item pool during the test, to cut down the unnecessary calculation; for another, this method, determines the number of test items according to the unit’s weighting, and it also pays close attention to the content balancing problems in the process of item selection. Figure 2 shows the system’s views of the Topic 2. For example, when Topic 2 is being tested, its general knowledge tests’ proportion takes 0.1 of total, while sub-unit Topic21 takes 0.4 of total tests, and sub-contents Concept5 takes 0.5 of total, so on and so forth, giving full consideration to multistage test contents. 3.2 CBIS-Based Item Selection Algorithm in Adaptive Testing 3.2.1 The CAT Testing Process Based on CBIS Item Selection Algorithm The CAT testing process based on CBIS item selection algorithm includes the following six steps, which is shown in the following figure 3: (1) Estimate testers’ initial ability level, to determine the test starting point; (2) The system selects and presentation of the item according to the estimation in the first step; (3) The testers answer the questions shown by the system;

Fig. 3. The process of CAT based of CBIS

The Content Balancing Method for Item Selection in CAT

179

(4) The system re-estimates testers’ ability level consulting the answers they gave, and meanwhile amends the item parameters; (5) According to the requirement of content balancing and the exposure rate of every items, determine the scope and select the item; (6) Repeat step (2) and (5) until the terminal conditions are fulfilled, and then terminate the tests. 3.2.2 The Test Starting Point and Initial Ability Level Estimation The testers need to take a pre-test before they sit the formal ones, so as to estimate their initial ability level and determine their test starting point. The distribution range of the testers’ ability level is from zero to ten, which is initially divided into four grades, respectively, (0, 2.5], (2.5, 5.0], (5.0, 7.5], and (7.5, 10]. Select one item with the highest discrimination (a=1), the medium difficulty coefficients (b=0.5), and guessing parameter(c=0.25), from the item pool to give testers a trial, in order to determine their ability’s attribution range. In this way, the testers’ test length can be cut off due to the estimation of their initial ability. 3.2.3 CBIS-Based Item Selection In CAT, item selection usually includes two phases: phase one is to determine the test content domain; phase two is to select items in the light of related controlling conditions. 1) The consideration of content balancing issue during the testing process. In content balancing design, it involves the upper bound and lower bound issues when selecting items from each test unit. Thereupon, when selecting test items, the following two formulas must be satisfied by the content balancing based adaptive testing: lk ≤ μ k ≤ uk . K

∑μ k =1

k

(3)

=L .

(4)

where lk and uk represent the upper bound and lower bound of selected item numbers respectively; μk refers to the item numbers selected from unit Ck ; k is the total units number; L represents the test length. During the testing process, select the candidate unit with the greatest weighting to be the first testing one by the weighting relations between units, and the formula is as follows (5). And then, under the circumstances of meeting the minimum testing item numbers of each unit, the items can be selected on the units weighting from large to small basis to continue the test, which has helped to solve the content balancing problems. Pi =

wi . ∑wj

where wi is the weighting of candidate unit Ci ;

(5)

∑w

didate units; Pi is defined as the probability of Ci .

j

is the weighting of all the can-

180

P. Lu et al.

2) Item exposure control. The calculation of exposure rate is shown in the following formula (6): the exposure rate of the largest item need to be controlled (e.g. ≤0.2). The control of item exposure rate not only influences the test validity, but also relates to utilization rate of the item pool. And one thing relates to exposure control is test coverage rate, as shown in the following formula (7). E= M

C=

∑I k =1

k

lk . N

(6)

(I k − 1)

L × N( N − 1)

.

(7)

where Ik is the exposure frequency of k items; L is the average test length; N is the number of testers; M refers to capacity of the item pool. 3) Item Selection. Based on the MI method and the weighting of item-relevant unit, the items can be selecting in accordance with the formula (8)：

CI i = w i × I i .

(8)

where Ii represents each item’s information content through MI method, which is concluded on the basis of current testers’ provisional estimated ability level θ’; wi represents the weighting of the item-relevant unit in the testing content. 4) The item difficulty coefficients adjustment. As the test goes on, the promptly adjustment to the difficulty coefficients b of each item is needed to fulfill test demands constantly, and to increase test accuracy. In CBIS method, the item difficulty coefficient b is determined by the initial item difficulty b0 and the information of its whole testing history, which is shown in the following formula (9): n

bi =

100 × b 0 + ∑ k × 1

100 + n

1 θj' 10 .

(9)

Among which, n is the total numbers of Item-i appeared in testing history; b0 is the initial item difficulty determined by the domain experts before entered into the item pool; θj’ is the ability score every tester owned when testing item i, with its scope from zero to ten integers; k is a variable, when the tester gives correct answer to item I, then k=0, or else k=1. It can be reflected from the above formula that the item difficulty coefficients b will amend its difficulty actively according to the testers’ test conditions, and the range of b is from zero to one. 3.2.4 Ability Estimation and Test Termination Throughout the test, the testers’ ability level θ need to be amended on and on, and each time an item is tested, their ability level needs to be adjusted correspondently to get the precise result. This CBIS method pre-establishes the biggest test length is L (50 in the experiment) and the threshold value is ε (0.01), and when the difference value of the two consecutive estimated tester’s ability level is below ε, or the biggest test length L is reached, then the test should be terminated. And the tester’s temporary ability level θ’ calculated at the terminate moment would be taken as his final ability level θ.

The Content Balancing Method for Item Selection in CAT

181

4 Case Study In this experiment, we choose Mathematics Item pool of Junior High School, and the item pool has 1000 items. Table 1 shows the content characteristics.

Content domains

Table 1. Item distribution

Algebra Statistics Geometry

Conceptual knowledge 94 62 85

Knowledge Categories Procedural Knowledge 181 177 158

Controlled Knowledge 92 43 108

All of Items belong to the same content domain in the item pool, and the whole course is divided into eight units, each of which contains one hundred and five relevant items. In the study, the content domain and the Knowledge Categories of each item were indexed when input into the item pool, and were assigned the appropriate weight. Each Item has limited content balancing and exposure rate. All the items are choice questions with four options, but only one option is correct, so the guessing parameter c is 0.25.In the experiment, two hundreds testers are chosen. They are divided into two groups, and each group contains 100 testers. The First group was tested according to MI Question selection items, while the second group was tested according to CBIS, and their ability levels range from zero to ten in integer. Initially, the testers’ ability levels are generated randomly by function R (0, 10). And in order to guarantee the test pool’s security, the exposure rate of each item should be controlled within 0.2. The regular writing test would be imitated, in which all the testers sit the test at the same time within the same time-scale. Figure 4, figure 5 and figure 6 are the comparisons of CBIS-based and MI-based item selection algorithms on the item pool’s utilization rate, item content balancing, and item exposure rate.

Fig. 4. Pool usage rate between CBIS and MI

182

P. Lu et al.

Figure 4 is the experiment results of the item pool’s utilization rate based on these two different algorithms. From it, a conclusion can be drawn that the item pool’s utilization rate of CBIS-based algorithm is becoming larger and larger with the ever-increasing testers’ number; and when the one hundred testers have all been tested, the whole pool’s utilization rate reaches 0.83, which makes clear that this algorithm has selected different difficulty items from different units to different testers. Meanwhile, the item pool’s utilization rate of MI-based algorithm is also increasing, but with rather small degree. After the one hundred testers have all finished their tests, the utilization rate is forty one percent, showing that part of the items are repeated on and on to different testers. Therefore, it can be concluded from the above test results that in terms of the item pool utilization rate, the CBIS-based item selection algorithm is much better than the MI-based one. Hence, the item pool utilization rate of CAT system basing on CBIS algorithm is much higher, and the cost to develop and maintain the pool would be smaller.

Fig. 5. Content Balance distributions for CBIS and MI

Figure 5 is the item coverage number related to each test unit (all together eight units), after all the testers having finished their tests. Among them, the item coverage rate of CBIS-based algorithm takes large proportion in every unit, and in a very even distribution. The unit that contributes the largest item number is unit seven, with one hundred and twelve items; and the unit contributing the smallest item number is unit four, with only ninety-eight items. So it can be concluded that the CBIS-based item selection algorithm executes a great control of the entire test theme on content balance during the test; the average testing item of each theme is one hundred and four. While the MI-based one is rather bad at this aspect: the item number selected from some units is too big, and some is too big. For instance, in this experiment, the testing item numbers from unit2、 unit3、 unit6 and unit8 are all below 25, while the items number from unit4 is one hundred and twenty-two, which proves that the item utilization from the former units is comparatively low, and the testing items for most testers would be selected from unit4. As a result, it may greatly influence the test security, and a great deal of testers may get to know the test items before hand.

The Content Balancing Method for Item Selection in CAT

183

Fig. 6. Exposure distributions for CBIS and MI

The figures in figure 6 shows that the CBIS-based algorithm adopts necessary method to control the item exposure rate within 0.2, and meanwhile keeps low test coverage rate to avoid the problem of high exposure rate of items. And when using MI-based item selection algorithm to test, the exposure rate of part items is too high, some even as high as 0.83, which has posed serious security problems to the item pool. Testers may have possibly gotten to know some of the items before hand, and then study relevant contents. They would continuously select items with high difficulty coefficients when sitting the tests to get better results as well as to cheat the test systems. These items, however, do not in accordance with testers’ real ability levels, which in turn, may affect the whole item pool’s security and the test outcomes.

5 Conclusion and Future Work This paper brought forth a content balancing-based item selection algorithm CBIS, which has given full consideration to the key issues in adaptive testing, such as the content balance of testing, item exposure rate, and the item pool utilization rate etc. throughout the testing process. From the above experiment, it can be drawn that the CBIS-based item selection algorithm’s indexes are much better in every aspects compared with MI-based one, with rather low item exposure rate, high item pool utilization rate, and full consideration of content-balanced problems of testing. Moreover, compared with the MI-based algorithm’s calculation of each item’s maximum information when selecting items, CBIS-based one only need to calculate the maximum information of items related to testers, hence, the time complexity is greatly reduced, and the realization is much easier. Though compared with MI-based algorithm, the CBIS-based one has made progress in many key issues. But it still needs to improve in some aspects; the most important is the response time on test items. Next step, the main focus would be the further study of the item response times, item exposure rate control and the further consideration of the item discrimination parameters in the testing process; and at the same time, to think over the organization of the item pool with the view to enhance the speed of searching and calculation of item related information to shorten the time.

184

P. Lu et al.

References 1. Conejo, R., Guzmán, E., Millán, E., Trella, M., PérezdelaCruz, J.L., Ríos, A.: SIETTE: a Web-based tool for adaptive testing. J. Artif. Intell. Educ. 14, 29–61 (2004) 2. Lord, F.M.: Some test theory for tailored testing. In: Holtzman, W.H. (ed.) Computer assisted instruction, testing and guidance, pp. 139–183. Harper and Row, New York (1970) 3. Wainer, H., Mislevy, R.: Item Response Theory, Item Calibration and Proficiency Estimation. In: Wainer, H. (ed.) Computerized Adaptive Testing: A Primer, pp. 65–102. Lawrence Erlbaum Associates Publishers, Hillsdale (1990) 4. Cheng, Y., Chang, H.-H.: The maximum priority index method for severely constrained item selection in computerized adaptive testing. British Journal of Mathematical and Statistical Psychology 62, 369–383 (2009) 5. Birnbaum, A.: Some Latent Trait Models and Their Use in Inferring an Examinee’s Mental Ability. In: Lord, F.M., Novick, M.R. (eds.) Statistical Theories of Mental Test Scores. Addison-Wesley, Reading (1968) 6. Hambleton, R.K.: Principles and Selected Applications of Item Response Theory. In: Linn, R.L. (ed.) Educational Measurement. MacMillan, New York (1989) 7. Jian-quan, T., Dan-min, M., Xia, Z., Jing-jing, G.: An Introduction to the Computerized Adaptive Testing. US-China Education Review, USA, Serial No.26 4(1) (January 2007) ISSN1548-6613 8. Gershon, R.C.: Test Anxiety and Item Order: New Concerns for Item Response Theory. In: Wilson., M. (ed.) Objective Measurement: Theory into Practice, ch. 11, vol. 1. Ablex, Norwood (1992) 9. Thissen, D., Mislevy, R.J.: Testing algorithms. In: Wainer, H. (ed.) Computerized adaptive testing: A primer, 2nd edn., pp. 101–133. Erlbaum, Mahwah (2000) 10. van der Linden, W.J.: Bayesian item selection criteria for adaptive testing. Psychometrika 63, 201–216 (1998) 11. Leung1, C.-K., Chang, H.-H., Hau, K.-T.: Computerized adaptive testing: A mixture item selection approach for constrained situations. British Journal of Mathematical and Statistical Psychology 58, 239–257 (2005) 12. Stocking, M.L., Swanson, L.: Optimal design of item banks for computerized adaptive tests. Applied Psychological Measurement 22, 271–279 (1998) 13. Huang, S.X.: A Content-Balanced Adaptive Testing Algorithm for Computer-Based Training Systems. In: Lesgold, A.M., Frasson, C., Gauthier, G. (eds.) ITS 1996. LNCS, vol. 1086, pp. 306–314. Springer, Heidelberg (1996) 14. Guzmà, E., Conejo, R.: A model for student knowledge diagnosis through adaptive testing. In: Lester, J.C., Vicari, R.M., Paraguaçu, F. (eds.) ITS 2004. LNCS, vol. 3220, pp. 12–21. Springer, Heidelberg (2004) 15. Cheng, Y., Chang, H., Yi, Q.: Two-phase item selection procedure for flexible content balancing in CAT. Applied Psychological Measurement 31, 467–482 (2007) 16. Chang, H., Ying, Z.: A-stratified multistage computerized adaptive testing. Applied Psychological Measurement 20, 213–229 (1999) 17. Parshall, C.G., Kromrey, J.D., Harmes, J.C., Sentovich, C.: Nearest neighbors, simple strata, and probabilistic parameters: An empirical comparison of methods for item exposure control in CATs. Paper presented at the Annual Meeting of National Council on Measurement in Education, Seattle, WA (2001) 18. Chang, H., Qian, J., Ying, Z.: A-stratified multisage CAT with b-blocking. Applied Psychological Measurement 25, 333–341 (2001)

The Formative Evaluation's Impact on Online Learning Mei Pu and Lu Wang Capital Normal University,Educational Technology, Beijing, China [email protected], [email protected]

Abstract. Formative evaluation has so much impact on learning, especially the online learning. There are many researches on the formative evaluation of online learning, but most of them focus on the theory or the system development, and the empirical researches on this aspect are still very limited. We proved the formative evaluation’s impact on the online learning by studying on the attention of the formative evaluation got from the learners, the course participation, the learner autonomy, and so on. We used the case study method, chose the education master's professional required course "Learning science and technology" and one of the learners to be the cases. In this paper, we will introduce the course, the formative evaluation, the study method and the relationship between the formative evaluation and online learning. Keywords: Online learning, Formative evaluation, Case study, Learner autonomy, Course participation.

1 Introduction Evaluation has sufficiently powerful effects on learning to be the course [1]. The formative evaluation became an integral part of learning evaluation since it had been proposed by Scriven in the "evaluation methodology" in 1967. Formative evaluations are designed for the purpose of giving feedback on performance and suggestions for improvement, and are intended to promote students' learning [2][3]. Formative evaluations that provide timely, relevant and supportive feedback (not just grades) can contribute to improved learning outcomes [4]. With the rapid development of information technology, the education which is based on the internet is the new direction of the distance education development [5]. Because of the characteristics of the environment, technology, and mode, online learning depends much more on formative evaluation than traditional learning. From the review, we can know that more focus on learning platforms and resources' building, the attention of learning evaluation is not enough attention [6]. In the online learning, the teaching organization structure of the students, teachers and administrators are loose, learning is not limited by time and place. Obviously, the traditional formative evaluation has been difficult to adapt to the needs of online learning [7]. Computers have done much to the online learning formative evaluation.The research about the computer support formative evaluation become more and more [8]. In fact, it's facility to conduct formative evaluation in the online learning than the traditional learning. The computer and network can keep the study X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 185–191, 2010. © Springer-Verlag Berlin Heidelberg 2010

186

M. Pu and L. Wang

record, make the formative evaluation feedback more quick, and make place facility, and so on [9]. The formative evaluation means a lot to the online learning. The experience of the online learning let me know the importance of the formative evaluation. Many scholars and experts have done many researches on the formative evaluation of online learning, but most concentrated in the theoretical study and the development of evaluation system, the empirical studies of formative evaluation on online learning are still small. To verify the formative evaluation's impact on online learning, and also to provide more evidence to the study of formative evaluation, we choose a course and one learner to conducted the case study.

2 Background In March 2009, the author participated in two courses offered by Professor Wang in the Virtual Learning Community of Capital Normal University, and served as a teaching assistants for one of them from March to July. 2.1 Virtual Learning Community of Capital Normal University Virtual Learning Community of Capital Normal University (VLCCNU) was developed by the team of professor Wang in 2000. The VLCCNU provides different tools for teachers and students [10]. The tools for learners such as my classroom, learning forums, virtual chat, personal space, scheduling, bookmarks management, and related downloads. And there are also a variety of support tools for teachers' teaching, management, and counseling organizations, such as the management of students, understanding the students, resource management, log management, operations management and course management, etc. The VLCCNU has been used for more than a decade, and more than 1,000 students had the experiences of learning through the VLCCNU. The students include primary, secondary, university and graduate. Even now, there are at least 2 courses to be conducted on the VLCCNU each term. At the beginning of the term, the learners should choose the course in the VLCCNU, and wait for the permission of the teachers. If the teachers had give the permission, then the students can participate in the course, learning through the resources, posting, finishing the task, having group meetings, and so on. 2.2 The Study of Science and Technology "The study of science and technology" is an education master's professional required course of the Education Technology Department. All the learners of the education master are teachers or administrators from the school, they are not full-time students. This course began from March 2009 to July 2009. Eighteen students had participated in the course, including nine education masters, eight graduate students, and one undergraduate student. The learners came from different provinces, they can't learn together in the class because of the work, the family, and so on. So they all learned, discussed and communicated through the VLCCNU. The forms of the courses activities include debates, seminars, group learning, role-playing and so on. At the beginning of the activities, teacher will give the tasks at the forum, and upload some materials about the activities. Learners read the materialson

The Formative Evaluation's Impact on Online Learning

187

line or download them, organize or participate the group meetings and discussions, help the others to solve the problems, share learning experiences, completed the tasks, and so on. According to the principle of heterogeneous group, learners are divided into two groups, each group was responsible for different tasks at each step. The group leader and his team conduct meetings regularly, complete the task with the team, and upload the work results. Each team can check the other team's conference records, the task results, and conducted the group's evaluations with the evaluation criteria developed by the other. 2.3 The Evaluation The total scores of the course include three parts, the formative evaluation, the group learning, and the paper. The formative evaluation accounts for 30% of the total, while the group learning accounts for 30%, and the last 40% is for the paper. The formative evaluation was conducted every three weeks by the course administrator, and the evaluation criteria was developed by the teacher at the beginning of the course. The formative evaluation scores are the total of the "basis points", the "post points" and the "reply points". If the learners want to get the "basis points" and "post points", their posts must be greater than or at least equal to the average in unit time (evaluation cycles), and the post correlation must be greater than or equal to 0.6. The post correlation refers to the relationship between the post content and the course. If the posts have a closer relationship with the course, the post correlation will be high, otherwise low. The "reply points" mainly focus on the attention the leaners' posts get. If one learner's posts get many replies, it means that the posts got much attention from the teacher and other learners, it also means that the posts give much to the other learners and the course, so he can get high points, vice versa. In this formative evaluation criteria, the "basis points" and "post points" are the evaluation to learners' attitude toward learning, and the "reply points" is the evaluation to learners' contribution. In the online learning , the learners' attitude is very important. The knowledge goals constructed by the active and passive learners are very different [7].

3 Study Methods The method of the study is case study. Case study, also known as case-finding for a particular individual and unit, or topicand phenomenon [11]. Case study research is one of many ways of doing social science research. A case study can be explanatory, exploratory or descriptive [12]. Case study may be targeted at individuals, groups, organizations, events, or a particular set of issues [11]. As Bent Flyvbjerg [13] said, the case study is a necessary and sufficient method for certain important research tasks in the social sciences, and it is a method that holds up well when compared to other methods in the gamut of social science research methodology. The data of the study came from the VLCCNU, and the object was the education master's professional course of the Department of Educational Technology, which was called "Learning Science and Technology".

188

M. Pu and L. Wang

In this research, several sets of quantitative data are collected, including the number of the posts, the topics, the readings and replies of the posts. All the data were analyzed with Excel 2003. The charts of the data will be shown to make the them clearer. Also the research has analyzed an individual learner, including the performance, the formative evaluation records, and the feeling of the learner.

4 Discussion 4.1 The Learners Pay Much Attention to Formative Evaluation As of July 2, the total posts of the course forum are1203, and the topics are 155. The average replies of each topicare 6.75, while the average readings are 31.14. In all the 155 topics, there are three topics about the formative evaluation reports. The average replies of the three formative evaluation reports' posts are 11.67, and the average readings are 48.33. This will be shown in figure 1:

Fig. 1. The average readings and replies of all topics and the formative evaluation

From figure 1 we can know that, compared to all the topics, the average replies and readings of the formative evaluation reports are much higher than all the topics. Some learners even have posted the topics about their feeling of the formative evaluation, so we can learn that the learners keep a watchful eye on the formative evaluation. The formative evaluation of this course which got so much attention has some relationship with the weights allocation of the course evaluation. In China, the graduate students must get at least 75 points to pass the exam, therefore, the formative evaluation accounts for 30% of the total score is very important. If somebody havn't got the formative evaluation points, he can't pass the course even though how much he got from the group learning and papers. The leaners won't pay much attention to the formative evaluation if it only accounts for a small part, however, if it takes up for a large percentage, they will ignore the group learning and papers. The evaluation's account is a key factor of the course, so the teachers should formulate an appropriate evaluation criterion at the beginning of the course. 4.2 Formative Evaluation Influence Learner Autonomy Learning autonomy requires learners to take their responsibilities, control the study, attribute their success or failure to their work and learning strategies rather than other factors which won't be controlled by themselves [14].

The Formative Evaluation's Impact on Online Learning

189

Many learners don't have good autonomy at the beginning of this course, figure 2 is the reply of one learner who had checked the second formative evaluation reports:

Fig. 2. The reply of one learner

The reply can be translated like this: I hope you can understand my feeling. The feeling that I took time to participate in the study and the discussion, replied every post which I can understand, but finally the points of formative evaluation was zero. I have remembered the rules you made for the evaluation, and I will abide by the rules strictly from now on. The evaluation records show that wangyihan got no points at the first two formative evaluation. He never posted at the first learning phase, that means he haven't participate in any discussion. From the reply of wangyihan (Figure 2), we can know that he made some efforts in the second phase of the study, such as answered the question, discussed with the others, and so on. But that does not mean he had done a good job, he even hadn't finish the study tasks. Although he had made 6 posts, but it was far from the standard, so the points is still zero. At the beginning of the study, wangyihan did not assume responsibility for his own learning, and accomplish the tasks, he even attributed the failure to the factors which were not under his control, so we can learn that the learning autonomy of wangyihan was very low at the first two phases. After the second formative evaluation, wangyihan took much more time to study, participate the discuss, and finish the tasks. Although figure 2 shows that wangyihan had some misunderstanding of formative evaluation, but it is undeniable that formative evaluation had help wangyihan to understand and assume his learning responsibility. In the record of the third formative evaluation report, we can see that wangyihan had gotten 4 points. Comparing with the previous two stages, it's a great progress. Through the analysis above we can see that the formative evaluation have impacted on the learner autonomy, help the learners to assume responsibility, complete the learning task, and then increasing the efficiency to learn. 4.3 Formative Evaluation Influence the Course Participation Course participation is an important factor to judge the community learning effects. As of July 2, the course administrators have conducted 3 formative evaluations. The readings and replies' points of the three topics (the formative evaluation) posted by the course administrators will be shown in figure 3:

190

M. Pu and L. Wang

Fig. 3. The reading and reply of each formative evaluation reports

The change of the daily posts amount at the forum will be shown in figure 4:

Fig. 4. The changes of every stage's average daily posts amount

As can be seen from figure 3, the first formative evaluation had gotten little attention, but this had been changed after the second evaluation. The attention which leaners payed to the formative evaluation increased rapidly, and had been stable after the third formative evaluation. The learning of a new phase is after last formative evaluation, just like that before the second phase of learning, the first formative evaluation must has been conducted. The administrators will publish the results of the evaluation, learners can express their views and opinions. From figure 4 we can know that the amount of daily posts of the first phase is minimum, and the amount has increased after the first formative evaluation. The biggest increase was after the second formative evaluation, and the third formative evaluation reports shows that the third stage's first topics and the replies are nearly the sum of the first two phases. So we can infer that the average daily posts of the first phase are so few, part of the reason is that there is no formative evaluation before. And the high attention which the second formative evaluation has gotten impacted much on the study of the third phase, that's why the average daily posts of the third phasewere so many. The high attention the learners payed to last formative evaluation made them spend much more time on the study, and the daily posts had increased. Then the daily posts will be steady when the attention which the learners pay to the formative evaluation became steady. The formative evaluation make the student pay more attention to the study, and increase the course participation.

The Formative Evaluation's Impact on Online Learning

191

5 Conclusions From the study we can know that the learners pay much attention to the formative evaluation, and the formative evaluation has some impact on the learner autonomy and the course participation, also learning outcome. So the teachers and the course administrators should conduct the formative evaluation as good as they can. The formative evaluation’s positive impact on the learning are based on the perfect evaluation criteria and implementation, otherwise it would be counterproductive. Therefore, both the evaluation criteria and implementation are very important. The impact of formative evaluation to learning is multifaceted, we only verify a small fraction of them. We hope to expand the scope of the study, and provide much more well-developed theory and practice for the online learning and formative evaluation.

References 1. Ramsden, P.: Learning to Teach in Higher Education. Routledge, London (1992) 2. Rolfe, I., McPherson, J.: Formative assessment: how am I doing? J. Lancet 345, 837–839 (1995) 3. Relan, A., Uijdehaage, S.: Web-based assessment for students’ testing and self-monitoring. J. Acad. Med. 76, 551 (2001) 4. Gipps, C.V.: What is the role for ICT-based assessment in universities? J. Stud. High Educ. 30, 171–180 (2005) 5. Li., H., Zhang, C.: To realize formative evaluation in the web-based course by flash. J. Modern Educational Technology 13(2), 63–66 (2003) 6. Zhu, F.: The study on the web-based formative evaluation. Yangzhou Universit. (2005) 7. Zhi, X.: The research of the online teaching’s formative assessment. J. Journal of ZhaoQing University 127(2), 22–24 (2006) 8. Dalziel, J., ScottGazzard: Next generation computer assisted assessment software: the design and implementation of WebMCQ. In: Fifth International CAA Conference 2001 (2001) 9. Zhang, J., Guozhen, J.: Analysison the effectiveness and timeliness of web-based formative evaluation. J. Education Research 13(2), 105–109 (2007) 10. Wang, L.: The Principle and Application of the Virtual Learning Community. Higher Education Press (2004) 11. Gall, M.D., Borg, W.R., Gall, J.P.: Educational research: An introduction, 6th edn. Longman, White Plains (1996) 12. Introduction to case study, http://74.125.155.132/scholar?q=cache:aH-RGBeGja8J:scholar. google.com/+Introduction+to+case+study&hl=zh-CN&as_sdt=2000 13. Flyvbjerg, B.: Five Misunderstandings About Case-Study Research. J. Qualitative Inquir. 12, 219–245 (2006) 14. Guo, Q.: Use the formative evaluation to promote the learner autonomy of the College English learning. J. Journal of Xi’an International Studies University 12(2), 66–68 (2004)

Psychological Perspectives on Social Behaviors of Chinese MMORPG Players Ge Qian Humanities College, Shanghai University of Finance and Economics 200433 Shanghai, China [email protected]

Abstract. This paper presents results of a psychological study on social behaviors of Chinese MMORPG players. Apart from basic demographic information, four types of social behavior were selected for the study: (1) respondents have told personal secrets to their MMORPG friends which they have never told their real-life friends; (2) respondents agreed that their MMORPG friends were better than their real-life friends; (3) respondents have physically dated someone they met in an MMORPG; (4) respondents considered themselves addicted to the MMORPG. The questionnaire used in this research was served to 500 subjects (347 effective) in Shanghai, China. It is found that a typical MMORPG avatar may be a high-school student, a computer or internet engineer in his 20’s, a stay-at-home mom and dad. But unlike the classic research by Nick Yee, the differences between the gender comparison or the age comparison of the above four different types of MMORPG behaviors are not statistically significant. Keywords: Massively multi-user online role-playing game; Social behavior; Addiction; Chi-square test.

1 Introduction Every day, millions of users interact, collaborate, and form relationships with each other through avatars in online environments known as Massively Multi-User Online Role-Playing Game (MMORPG). These online environments offer tantalizing glimpses of how millions of avatars interact on a daily basis outside of a laboratory setting and what users derive from that experience. 1.1 Literature Review Since the interactions between MMORPG players are real, even if the environments are virtual, psychologists and sociologists are able to use MMORPGs as tools for academic research. Sherry Turkle, a clinical psychologist, has conducted interviews with computer users including game-players. Turkle found that many people have expanded their emotional range by exploring the many different roles (including gender identities) that MMORPGs allow a person to explore [1]. Nick Yee has surveyed more than 35,000 MMORPG players over the past several years, focusing on psychological and sociological aspects of these games. Recent X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 192–202, 2010. © Springer-Verlag Berlin Heidelberg 2010

Psychological Perspectives on Social Behaviors of Chinese MMORPG Players

193

findings included that 15% of players become a guild-leader at one time or another, but most generally find the job tough and thankless;[2] and that players spend a considerable amount of time (often a third of their total time investment) doing things that are external to game play but part of the meta game [3]. Many players report that the emotions they feel while playing an MMORPG are very strong, to the extent that 8.7% of male and 23.2% of female players in a statistical study have had an online wedding [4]. Other researchers have found that the enjoyment of a game is directly related to the social organization of a game, ranging from brief encounters between players to highly organized play in structured groups [5]. In a study by Zaheer Hussain and Mark D. Griffiths, it was found that just over one in five gamers (21%) said they preferred socializing online to offline. Significantly more male gamers than female gamers said that they found it easier to converse online than offline. It was also found that 57% of gamers had created a character of the opposite gender, and it is suggested that the online female persona has a number of positive social attributes [6]. Richard Bartle classified multiplayer RPG-players into four primary psychological groups. His classifications were then expanded upon by Erwin Andreasen, who developed the concept into the thirty-question Bartle Test that helps players determine which category they are associated with. With over 200,000 test responses as of 2006, this is perhaps the largest ongoing survey of multiplayer game players [7]. In World of Warcraft, a temporary design glitch attracted the attention of psychologists and epidemiologists across North America, when the “Corrupted Blood” disease of a monster began to spread unintentionally—and uncontrollably—into the wider game world. The Center for Disease Control used the incident as a research model to chart both the progression of a disease, and the potential human response to large-scale epidemic infection [8]. 1.2 MMORPG Study in China As of December 2007, the number of online game players in China had reached 40 million, up 23% growth in 2006. According to the Chinese online game industry conference 2007, China's online game market net revenue in 2007 was $1.47 billion USD, up 61.5 % growth in 2006. And according to CNNIC (2008) survey, the use rate of online game by Chinese natives accounts for 59.3%, even higher than e-mail use which is 56.5%. Although the above data are all about online game, different from the online card game or chess game, the MMORPG is the most important part of online game. Table 1 shows the top 10 MMORPG in mainland China 2007. According to an online survey of Sina.com and QQ.com, which based on 9,915 college students in 2006, 30% people play online game every day, 29% often play, 27% play sometimes, and 14% never play. Moreover, 55% students have skipped classes because playing online game. Nowadays, the MMORPGs have tremendous economic and social influence to the Chinese society, but the academic research of MMORPGs is very limited, especially in psychology. Past researches of online games by the China mainland scholars suggested that China’s online game industry should learn from South Korea and train Chinese own game developers [9, 10, 11]. Some scholars studied the addiction of online games among Chinese adolescents [12, 13, 14, 15] and some stated how to use online games in the field of education [16, 17].

194

G. Qian Table 1. Top 10 MMORPG in mainland China 2007 Rank 1 2 3 4 5 6 7 8 9 10

Name of the game Magic Journey to the West ZT online World of Warcraft Audition Crazy racing Kart rider MIR Heaven Dragon The Eight Episode ZX online MY online Street Basketball

2 Method Data were gathered in Shanghai, China. A total 500 pieces of questionnaire were distributed in middle school classrooms, university classrooms and dormitories, offices, computer classroom in adult education institutions, computer classroom in the college of elderly and retired, internet café and McDonald's, 347 pieces of questionnaire were back and the response rate was 69.4%. The questionnaire used in this research combines the scales of Nick Yee [18], Kimberly Young [19] and Xue Qiang [20]. In the questionnaire, survey items were used to gather responses to basic demographic information: gender, age, marital status, occupational status, hours of usage per week, and whether the user participated with a family member or romantic partner. And four types of social behavior were particular selected for the study: (1) respondents have told personal secrets to their MMORPG friends which they have never told their real-life friends; (2) respondents agreed that their MMORPG friends were better than their real-life friends; (3) respondents have physically dated someone they met in an MMORPG; (4) respondents considered themselves addicted to the MMORPG. This study used SPSS software for data processing, and the main statistical analysis method used in this paper is Chi-square test.

3 Results 3.1 Basic Information The majority of respondents were male (68.0%), the average age of the respondents was 24.77, the median was 23, with a range from 13 to 66. Table 2 shows that female players were significantly older than male players. It is found in this survey that a typical MMORPG avatar may be a high-school student, a computer or internet engineer in his 20’s, a stay-at-home mom and dad. Table 2. Age distribution by gender 13-17 18-22 23-29 30-40 >40 Total Male 46 69 65 48 8 236 Female 7 22 37 30 15 111

Psychological Perspectives on Social Behaviors of Chinese MMORPG Players

3.2 The Gender and Age Comparison of the MMORPG Behavior

Ⅰ

195

Ⅰ

MMORPG behavior here refers to the behavior that the MMORPG players have told personal issues or secrets to their MMORPG friends which they have never told their real-life friends. Table 3 show age distribution by gender of the MMORPG behavior ; Table 4-8 show the results of Chi-square test on gender comparison of the MMORPG behavior ; and Table 9-10 show the results of Chi-square test on age comparison of the MMORPG behavior .

Ⅰ

Ⅰ

Ⅰ

Table 3. Age distribution by gender (MMORPG behavior

Ⅰ)

13-17 18-22 23-29 30-40 >40 Male 9 19 12 8 1 Female 4 9 13 7 4 Table 4. Chi-square test on gender comparison of the MMORPG behavior

Male Female Total Chi-square test

Ⅰ(age: 13-17)

Occurrences No Occurrences Total 9 37 46 4 3 7 13 40 53 chi-square = 2.83 (corrected); 1 df; p > 0.05

3.3 The Gender and Age Comparison of the MMORPG Behavior

Ⅱ

Ⅱ

MMORPG behavior here refers to the behavior that the MMORPG players agreed that their MMORPG friends were comparable to or better than their real-life friends. Table 11 show age distribution by gender of the MMORPG behavior ; Table 12-16 show the results of Chi-square test on gender comparison of the MMORPG behavior ; and Table 17-18 show the results of Chi-square test on age comparison of the MMORPG behavior .

Ⅱ

Ⅱ

Table 5. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 19 Female 9 Total 28 Chi-square test

Ⅰ(age: 18-22)

No Occurrences Total 50 69 13 22 63 91 chi-square = 1.40; 1 df; P > 0.05

Table 6. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 12 Female 13 Total 25 Chi-square test

Ⅰ

Ⅰ(age: 23-29)

No Occurrences Total 53 65 24 37 77 102 chi-square = 3.54; 1 df; P > 0.05

196

G. Qian

Table 7. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 8 Female 7 Total 15 Chi-square test

Ⅰ(age: 30-40)

No Occurrences Total 40 48 23 30 63 78 chi-square = 0.53; 1 df; P > 0.05

Table 8. Chi-square test on gender comparison of the MMORPG behavior

Ⅰ(age: >40)

Occurrences No Occurrences Total Male 1 7 8 Female 4 11 15 Total 5 18 23 Direct probability method n = 23; 1 df; P = 0.32

3.4 The Gender and Age Comparison of the MMORPG Behavior

Ⅲ

Ⅲ

here refers to the behavior that the MMORPG players have MMORPG behavior physically dated someone they met in an MMORPG. Table 19 shows age distribution by gender of the MMORPG behavior ; Table 20-24 show the results of Chisquare test on gender comparison of the MMORPG behavior ; and Table 25-26 show the results of Chi-square test on age comparison of the MMORPG behavior .

Ⅲ

Ⅲ

Ⅲ

Table 9. Chi-square test on age comparison of the MMORPG behavior Occurrences 13-17 9 18-22 19 23-29 12 30-40 8 >40 1 Total 49 Chi-square test

No Occurrences Total 37 46 50 69 53 65 40 48 7 8 187 236 chi-square = 3.00; 4 df; P > 0.05

Table 10. Chi-square test on age comparison of the MMORPG behavior Occurrences 13-17 4 18-22 9 23-29 13 30-40 7 >40 4 Total 37 Chi-square test

Ⅰ(gender: male)

Ⅰ(gender: female)

No Occurrences Total 3 7 13 22 24 37 23 30 11 15 74 111 chi-square = 3.53; 4 df; P > 0.05

Psychological Perspectives on Social Behaviors of Chinese MMORPG Players Table 11. Age distribution by gender (MMORPG behavior

197

Ⅱ)

13-17 18-22 23-29 30-40 >40 Male 9 12 9 5 1 Female 2 6 9 4 2 Table 12. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 9 Female 2 Total 11 Chi-square test

No Occurrences Total 37 46 5 7 42 53 chi-square = 0.00 (corrected); 1 df; P > 0.05

Table 13. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 12 Female 6 Total 18 Chi-square test

Ⅱ(age: 18-22)

No Occurrences Total 57 69 16 22 73 91 chi-square = 0.50 (corrected); 1 df; P > 0.05

Table 14. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 9 Female 9 Total 18 Chi-square test

Ⅱ(age: 23-29)

No Occurrences Total 56 65 28 37 84 102 chi-square = 1.78; 1 df; P > 0.05

Table 15. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 5 Female 4 Total 9 Chi-square test

Ⅱ(age: 13-17)

Ⅱ(age: 30-40)

No Occurrences Total 43 48 26 30 69 78 chi-square = 0.00 (corrected); 1 df; P > 0.05

Table 16. Chi-square test on gender comparison of the MMORPG behavior Occurrences No Occurrences Total Male 1 7 8 Female 2 13 15 Total 3 20 23 Direct probability method n = 53; 1 df; p = 0.32

Ⅱ(age: >40)

198

G. Qian

Table 17. Chi-square test on age comparison of the MMORPG behavior Occurrences 13-17 9 18-22 12 23-29 9 30-40 5 >40 1 Total 49 Chi-square test

No Occurrences Total 37 46 57 69 56 65 43 48 7 8 187 236 chi-square = 1.92; 1 df; P > 0.05

Table 18. Chi-square test on age comparison of the MMORPG behavior Occurrences 13-17 2 18-22 6 23-29 9 30-40 4 >40 2 Total 23 Chi-square test

Ⅱ(gender: male)

Ⅱ(gender: female)

No Occurrences Total 5 7 16 22 28 37 26 30 13 15 88 111 chi-square = 2.63; 4 df; P > 0.05

Table 19. Age distribution by gender (MMORPG behavior

Ⅲ)

13-17 18-22 23-29 30-40 >40 Male 1 3 5 4 1 Female 1 1 3 2 1 Table 20. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 1 Female 1 Total 2 Chi-square test

No Occurrences Total 45 46 6 7 51 53 chi-square = 0.25(corrected); 1 df; P > 0.05

Table 21. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 3 Female 1 Total 4 Chi-square test

Ⅲ (age: 13-17)

Ⅲ (age: 18-22)

No Occurrences Total 66 69 21 22 87 91 chi-square = 0.31(corrected); 1 df; P > 0.05

Psychological Perspectives on Social Behaviors of Chinese MMORPG Players

Table 22. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 5 Female 3 Total 8 Chi-square test

Ⅲ (age: 23-29)

No Occurrences Total 60 65 34 37 94 102 chi-square = 0.09(corrected); 1 df; P > 0.05

Table 23. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 4 Female 2 Total 6 Chi-square test

199

Ⅲ (age: 30-40)

No Occurrences Total 44 48 28 30 72 78 chi-square = 0.03(corrected); 1 df; p > 0.05

Table 24. Chi-square test on gender comparison of the MMORPG behavior

Ⅲ (age: >40)

Occurrences No Occurrences Total Male 1 7 8 Female 1 14 15 Total 2 21 23 Direct probability method n = 23; 1 df; p = 0.47 Table 25. Chi-square test on age comparison of the MMORPG behavior Occurrences 13-17 1 18-22 3 23-29 5 30-40 4 >40 1 Total 14 Chi-square test

No Occurrences Total 45 46 66 69 60 65 44 48 7 8 222 236 chi-square = 2.95; 4 df; p > 0.05

Table 26. Chi-square test on age comparison of the MMORPG behavior Occurrences 13-17 1 18-22 1 23-29 3 30-40 2 >40 1 Total 8 Chi-square test

Ⅲ (gender: male)

Ⅲ (gender: female)

No Occurrences Total 6 7 21 22 34 37 28 30 14 15 103 111 chi-square = 1.43; 4 df; p > 0.05

200

G. Qian

3.5 The Gender and Age Comparison of the MMORPG Behavior

Ⅵ

Ⅵ

MMORPG behavior here refers to the behavior that the MMORPG players would consider themselves addicted to the MMORPG environment they participated in. Table 27 show age distribution by gender of the MMORPG behavior ; Table 28-32 show the results of Chi-square test on gender comparison of the MMORPG behavior ; and Table 33-34 show the results of Chi-square test on age comparison of the MMORPG behavior .

Ⅵ

Ⅵ

Table 27. Age distribution by gender (MMORPG behavior

Ⅵ

Ⅵ)

13-17 18-22 23-29 30-40 >40 Male 22 35 27 19 3 Female 5 10 16 13 6 Table 28. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 22 Female 5 Total 27 Chi-square test

Ⅵ (age: 13-17)

No Occurrences Total 24 46 2 7 26 53 chi-square = 0.57 (corrected); 1 df; P > 0.05

Table 29. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 35 Female 10 Total 45 Chi-square test

No Occurrences Total 34 69 12 22 46 91 chi-square = 0.19; 1 df; P > 0.05

Table 30. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 27 Female 16 Total 43 Chi-square test

Ⅵ (age: 23-29)

No Occurrences Total 38 65 21 37 59 102 chi-square = 0.03; 1 df; P > 0.05

Table 31. Chi-square test on gender comparison of the MMORPG behavior Occurrences Male 19 Female 13 Total 32 Chi-square test

Ⅵ (age: 18-22)

Ⅵ (age: 30-40)

No Occurrences Total 29 48 17 30 46 78 chi-square = 0.11; 1 df; P > 0.05

Psychological Perspectives on Social Behaviors of Chinese MMORPG Players Table 32. Chi-square test on gender comparison of the MMORPG behavior

201

Ⅵ (age: >40)

Occurrences No Occurrences Total Male 3 5 8 Female 6 9 15 Total 9 14 23 Direct probability method n = 23; 1 df; p = 0.34 Table 33. Chi-square test on age comparison of the MMORPG behavior Occurrences 13-17 22 18-22 35 23-29 27 30-40 19 >40 3 Total 106 Chi-square test

No Occurrences Total 24 46 34 69 38 65 29 48 5 8 130 236 chi-square = 2.13; 1 df; P > 0.05

Table 34. Chi-square test on age comparison of the MMORPG behavior Occurrences 13-17 5 18-22 10 23-29 16 30-40 13 >40 6 Total 50 Chi-square test

Ⅵ (gender: male)

Ⅵ (gender: female)

No Occurrences Total 2 7 12 22 21 37 17 30 9 15 61 111 chi-square = 2.21; 1 df; P > 0.05

4 Conclusion It is found in this survey that a typical MMORPG avatar may be a high-school student, a computer or internet engineer in his 20’s, a stay-at-home mom and dad. But unlike the classic research by Nick Yee, the differences between the gender comparison or the age comparison of the above four different types of MMORPG behaviors are not statistically significant. This may be because the questionnaire in the study of Yee is through Internet, but the questionnaire in my study is through pen and paper; and the subjects of Yee are Westerns, but my subjects are Chinese. Obviously, Westerns, rather than Chinese; or through Internet, rather than pen and paper, will be more likely to exhibit the uniqueness and diversity.

References 1. Turkle, S.: Life on the Screen: Identity in the Age of the Internet. Simon & Schuster, New York (1997) 2. The Daedalus Project, http://www.nickyee.com/daedalus/archives/001516.php

202

G. Qian

3. The Daedalus Project, http://www.nickyee.com/daedalus/archives/001535.php 4. The Daedalus Project, http://www.nickyee.com/daedalus/archives/000467.php 5. Nardi, B., Harris, J.: Strangers and Friends: Collaborative Play in World of Warcraft. In: Proceedings of the 20th Anniversary Conference on Computer Supported Cooperative Work, pp. 149–158. ACM Press, New York (2006) 6. Hussain, Z., Griffiths, M.D.: Gender Swapping and Socializing in Cyberspace: An Exploratory Study. Cyberpsychol. Behav. 11, 47–53 (2008) 7. Bartle Test of Gamer Psychology, http://www.gamerdna.com/quizzes/ bartle-test-of-gamer-psychology 8. CVG, http://www.computerandvideogames.com/article.php?id=131791 9. Chen, W.: The Inspiration of South Korea Games Support Policies. Publishing Research 2, 15–18 (2006) 10. Quan, W.: The Three Major Existing Problems and Solutions of China’s Online Game Industry. Friends of Editors 3, 64–66 (2004) 11. Yao, Y.: Comparison of China and Korean Online Game Industry. Friends of Editors 2, 67–69 (2005) 12. Zhang, L.: Behind the “great firewall”: Decoding China’s Internet media policies from the inside. Convergence: The Journal of Research into New Media Technologies 12, 271–290 (2006) 13. Liu, J.Y., Chen, S.H.: Investigation and Analysis on Internet Game Behaviors by Minor Students. Jiangxi Educational Research 11, 37–41 (2007) 14. Wang, B., Yu, H.B., Yang, S.: Online Game Addiction and Learning Burnout of University Students. Chinese Mental Health Journal 12, 841–844 (2007) 15. Yang, W.J., Zhou, Y.J.: The Relationship between the Type of Internet Addiction and the Personality Trait in College Students. Journal of Huazhong University of Science and Technology: Social Science Edition 3, 39–42 (2004) 16. Pu, Y.L.: On the Network Game as an Education for All-Round Development. Jianghan Tribune 5, 136–139 (2007) 17. Wang, Y.S., Bian, X.N., Zhang, P.F.: Education Network Games Open the Door to Learning for Human. E-education Research 9, 35–38 (2007) 18. Yee, N.: The Demographics, Motivations and Derived Experiences of Users of MassivelyMultiuser Online Graphical Environments. Teleoperators and Virtual Environments 15, 309–329 (2006) 19. Young, K.: Understanding online gaming addiction and treatment issues for adolescents. American Journal of Family Therapy 37, 355–372 (2009) 20. Xue, Q.: Demographics, Motivations, Addictions and Usage Patterns among Chinese College Student MMORPG Players. MSc thesis: Chinese University of Hong Kong (2008)

Research on the Adaptive Strategy of Adaptive Learning System Lian Bian1,3 and Yueguang Xie2,3 2

1 Ideal Institute of Information and Technology in NENU, JiLin ChangChun, 130024 School of Software in NorthEast Normal University (NENU), JiLin ChangChun, 130024 3 Engineering & Research Center of E-learning, JiLin ChangChun,130024 [email protected], [email protected]

Abstract. As the rapid development of information technology and the social sciences, adaptive learning becomes the development trend of distance education, while the adaptive learning system is an important research area nowadays. This study finds and proves mutual adaptation relationship among adaptive learning by the use of literature reading method, and puts forward the "Mutual Adaptation" idea applied in the system design. And the idea is applied to the system’s theoretical model and adaptive strategy. This study will benefit those who would like to design adaptive learning systems. Keywords: Adaptive Learning System; Mutual Adaptation; Theoretical Model; Adaptive Strategy.

1 Introduction and Problem Definition 1.1 Background “E-Learning can be viewed as an innovative approach for delivering well designed, learner-centered, interactive, and facilitated learning environment to anyone, anyplace, anytime by utilizing the attributes and resources of various digital technologies along with other forms of learning materials suited for open, flexible, and distributed learning environment”. [1]However, learners are individuals, and E-Learning research of resent years is slowly starting to take that into consideration, moving away from the “one-size-fits-all” approach of educational broadcasting. The move is faster in the research field, where many adaptive learning systems already take into account different learner features like goals/tasks, knowledge, background, hyperspace experience, preferences and interests. 1.2 Adaptive Learning System Adaptive learning system is a learning system providing learning support for individual characteristics because of the individual differences (due to people, due to time) in their learning processes. It is also essentially a kind of individualized learning support system providing a user personalization features to adapt to the user view which including X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 203–214, 2010. © Springer-Verlag Berlin Heidelberg 2010

204

L. Bian and Y. Xie

not only learning resources but also individualized learning process and strategies.[2] Until recent years, the adaptive learning system is still in the research stage, with the development of Internet and related new technologies, researchers started from a new perspective on system, and there are four categories which are adaptive hypermedia in education research, distributed adaptive learning system on web services, smart space for learning based on the next generation of web technology, e-learning grid. 1.3 Adaptive Techniques There are several adaptive techniques which are usefully employed in an educational environment. These methods include adaptive navigation, structural and historical adaptation and adaptive presentation. Navigation which attempts to guide the learner through the system by customizing the link structure or format according to a learner model includes guidance, link annotation, link hiding and link sorting. Structural adaptation which attempts to give the learner a spatial representation of the hyperspace environment includes maps, filters, fisheyes and indexes. Historical adaptation which is attempts to give a time context to the learner by adapting representations of the learner’s path includes landmarks, trails and footprints. Adaptation presentation which is the customization of course content to match learning characteristics specified by the user model, includes inserting/removing, sorting, altering and stretch-text. [3] From the above introduction, it is clear that the first, many researchers have designed and developed systems by a variety of frameworks or a variety of technology in order to achieve the adaptation of the system. The system model similar to the ITS model and most research focus on technology innovation to build the system, but without education theoretical support. At the same time there is no emphasis on learning as the center from the perspective of learners. Secondly, many researchers lack of understanding of learning theory, so affect the system's design. For example, how to identify personal characteristics of students especially non-intellectual factors; which characteristics reflects the students learning methods or even learning styles; how to approach the student to choose the right learning strategies; how to tap the information of learning rule; how these information make choice of different types of educational resources; how can we quantify the non-intellectual factors of user model, etc. Of course, the design which reflects the "personality" of the system on the theoretical level of teaching is precisely efforts in our next target.

2 Theoretical Analysis of Adaptive Strategy 2.1 Mutual Adaptation Relationship among Adaptive Learning In the learning process, individuals have the ability, background, learning style, learning objectives and other kinds of differences, even if the individual themselves, in the learning process, the state of knowledge are constantly changing. Learners can adjust and adapt to different environments. So as the needs of individual learners in a variety of learning activities should adapt to the existing learning environment, or the creation of an effective operating environment to promote the effective attainment of

Research on the Adaptive Strategy of Adaptive Learning System

205

learning by the way of meeting the various needs, such as evaluation, content, methods and so on. Therefore, the adaptive learning system can be here interpreted as learning environment. Thus learners to adapt the system can also be reflected in the learning diagnosis, selection and organization of learning content, presentation of learning content, mode of study. At the same time, system corresponds to adapt learners through the learning diagnosis, dynamic organization and presentation of the contents, mode of study to choose (Fig.1).

Fig. 1. Mutual Adaptation Relationship among Adaptive Learning

Thus, this study proposes the "Mutual Adaptation" idea. The idea means there is Mutual Adaptation relationship among adaptive learning, that is, technical support-based systems to adapt to learners at the same time, learners have to adapt to the system to make learning diagnosis, dynamic selection and organization of learning content, presentation of learning content, mode of study to choose which are environmental factors changed for them. So, in the study of adaptive learning system building methods, it should be the spirit of the "Mutual Adaptation" idea, from the perspective of learners to study respecting for individual differences, to study the psychological "learner adaptive system" assistance strategies and methods. [4] These strategies and methods applied to the design or implementation. 2.2 Theoretical Model of Adaptive Learning System in the Ideal State Modeling Major research area in this paper is the first class called adaptive hypermedia in education research. This study found that Mutual Adaptation relationship among adaptive learning, thus build a theoretical model of adaptive learning systems by the "Mutual Adaptation" perspective. Theoretical model of adaptive learning system in the ideal state is built mainly from the internal structure and adaptive strategy (Table 1).

206

L. Bian and Y. Xie

Table 1. Theoretical Model of Adaptive Learning System in the Ideal State

the Internal Structure

the Student Model

the Knowledge Space

Guidance Rules

Learning Diagnosis

Dynamic Organization and Presentation of the Contents Mode of Study to Choose

That describes information and data about an individual learner, such as knowledge status, learning style preferences, etc. The Student Model contains two distinct sub-models, one for representing the learner’s state of knowledge, and another one for representing learner’s cognitive characteristics and emotional features. That contains two sub-spaces. The first one, referred to as, the Media Space contains educational resources and associated descriptive information (e.g. metadata attributes, usage, attributes etc.) and the second, referred to as, the Domain Model contains graphs that describe the structure of the domain knowledge in-hand and the associated learning goals. The rules of matching between the Student Model and the Knowledge Space. These rules contain Concept Selection Rules which are used for selecting appropriate concepts from the Student Model to be covered, as well as, Content Selection Rules which are used for selecting appropriate resources from the Knowledge Space. Adaptive Strategy After using some editing of test measurement theory exercises to test students, system requires this test to measure student learning as much as possible, with minimum contents to diagnose the true ability of student. According to the results of learning diagnosis or learner's learning history, dynamic organize the most relevant learning content of learners currently learning. The presentation is choosing the most suitable presentation of contents based on cognitive style of learner. Based on the specific learning content, design navigation links to supports a variety of modes of study so that learner can choose by himself.

2.3 Adaptive Strategy As can be seen from the theoretical model, adaptive strategy refers to adaptation from the perspective of learner support in their studies which is also the adaptive behavior, mainly in the following three areas: learning diagnosis, dynamic organization and presentation of contents, mode of study to choose. System expresses "adaptive" by adaptive strategy, and learners make the feedback for "adaptive" of system is a manifestation of adaptive effectiveness. Learners can be more quickly accepted, and better to learn new content, which is the adaptability of

Research on the Adaptive Strategy of Adaptive Learning System

207

learners to achieve the highest value, be that the system to adapt to individual learners to achieve the best. Thus, for the system, it is not just the technical aspects to achieve adaptive, more importantly, take full account of learners and adaptability factors in the design of the system, which is the "Mutual Adaptation" idea to emphasize.

3 Learning Diagnosis 3.1 Theoretical Analysis As the view of teaching, learning diagnosis uses scientific methods to help students identify the learning bias, mistakes, analyze its causes, and to take effective measures to optimize , correcting the course by the way of intellectual factors, non-intellectual factors, subject knowledge points. This study suggests that adaptive learning system should be to achieve the same process of diagnosis, and learning diagnosis consists of four stages: diagnosis occurrence, producing results, searching for "causes", providing the decision-making. The system in order to consider the following dimensions at different stages (Fig. 2).

Fig. 2. The Four Stages of Learning Diagnosis

3.2 Design Management Management is divided into two parts: learner information, item management. Learner information management mainly administrate include the learner preferences, learner's state of knowledge, errors of knowledge records, historical activities, performance records which is related to the student model; Item management completes the establishment and maintenance of item, the questions entry, parameter of examination questions setting, examination questions query, modify, delete, and statistical functions. Teachers and administrators can carry out the examination questions entry and editing operations within the competence, but also some of the necessary parameters of the examination questions setting, such as kinds of questions, the amount of questions, points setting, the difficulty setting and guide words, sequence, answer forms,

208

L. Bian and Y. Xie

discrimination index, item response time constraints. They may also query information in the test database. Smart Quizzing Smart quizzing is mainly reflected in the user choice and item setting, principle of quizzing, optimization algorithms. The user choice and item setting are: first, learning can accord to learning needs of the user, according to types of courses, textbooks, chapters, teaching objectives, learning content keywords, knowledge points and other information to search examination questions. Secondly, they attribute information independently on the subject of a variety of settings, such as kinds of questions, the amount of questions, points setting, the difficulty setting and guide words, sequence, answer forms. System based on their requirements indicators to quiz. Principle of quizzing is system according to certain educational rules to optimize quizzing, such as format (fill-in, short answer and double-entry multiple-choice), simple to complex order. Optimization algorithms are three kinds: random selection method, backtracking heuristics, genetic algorithms. Marking and Assessment The main consideration of marking assessment is result form design. In order to support two types of goals, the result forms should be to distinguish. Corresponding to two kinds of presentation, with the raw scores or the percentage to express that learners has gone beyond the specified minimum acceptable performance levels; with the central tendency (such as the arithmetic mean, mode) to express that most of the other learners in the diagnosis, so learner could define his position. Result Diagnosis and Decision-making Result diagnosis and decision-making mainly based on class diagrams to search for precursor knowledge, based on the theory of IRT to diagnosis learners’ ability level, thereby making the next decision-making. It also need to understand learning ability and cognitive style of learners through analysis of the course of learning, let focus of learning content of most appropriate way to present learners. These require us to make comprehensive use of some new technologies, such as data mining. Application of the "Mutual Adaptation" Idea Analysis found that learners will have non-intellectual factors such as test anxiety, self-efficacy which affect their adaptation to the learning diagnosis. Therefore, the system imposes these areas assistance strategy through the following design. z

A clear distinction between three time differences of diagnosis occurred. Through the concrete, positive, concise language tips for learner’s hint; before diagnosis, an overview or review of prior learning; into the exam, let learners choose own their auxiliary means such as favorite music to relax, thereby improving the test anxiety. Also pay attention to, the time of diagnosis occurs should be consistent with the learner's cognitive patterns, reflected in two aspects: in the need to strengthen the knowledge, focusing on knowledge point to occur diagnosis; when

Research on the Adaptive Strategy of Adaptive Learning System

z

z

z

z

209

the learner requests the knowledge point to a diagnosis, the system can provide timely. Allows the learner to correctly understand the objective. Through adjustment of diagnosis forms, to distinguish between different types of tests. For example, using plain language to explain the meaning of target in the Language Tips; due to the different evaluation criteria, resulting in kinds of questions, the results show the distinction between two kinds of tests. Clear that the relevance of the content and safety of submission, the purpose is to eliminate the doubts of learners at this time. Inform the learner: Content, some features tell that the system should provide prompt teaching resources; Background, some features tell that system should prompt the creation of situations in teaching or psychological background; way, some features tell that system should provide teaching methods used in. [5] To give learners the power of personalization settings; Tips language reflects the humanity, harmony-oriented, in order to reduce and even eliminate negative emotions of learners, increase learning motivation; results demonstrate the concept of authentic assessment, using a reasonable form to display. The purpose is to allow the results generated by a tool of inspiration, rather than punishment. In addition, the number of questions need to control, not too many. The external strengthen. For instance, incentive mechanisms, the form of social comparison, displaying information on personal progress can be used.

4 Dynamic Organization and Presentation of the Contents 4.1 Theoretical Analysis As early as the early 20th century, Dewey had undertaken the research of teaching content knowledge. Shulman also creatively put forward the concept of teaching content knowledge. Content knowledge of teaching theory is divided into three parts: the knowledge of subject, the knowledge of student and the knowledge of study. And there is a complementary relationship in these three parts in student's cognitive processes. [6] This research suggests that the information of dynamic organization and presentation of the contents should also be involved in three: the information of subject, the information of student and the information of study. The information of subject mainly is academic characteristics, subject knowledge, instructional design, course structure, assisted learning resources and so on; the information of student mainly is level of knowledge, interests, preferences and so on; the information of study mainly is cognitive style or learning style, which refers to an individual has always been unique and lasting style of the information processing in cognitive performance and cognitive function in the organization. It includes individual perception, memory, thinking and other cognitive process differences, but also including individual attitudes, motivation and cognitive capacity of differences between personality formation and cognitive.

210

L. Bian and Y. Xie

4.2 Design Personalized Content Recommendation First of all, learners can choose their own interest in the learning content, so the system provides web-based courses multimedia material, item bank, case studies and other digital content for learners, courses, learning modules cluster (chapters, sections), knowledge-point cluster, and other self-learning content organization, text, pictures, photos , animation, audio, video and other multimedia learning resources. Secondly, in order to reflect "adaptive" of system, personalized recommendation technology becomes the core technology in dynamic organization and presentation of the content. At present, the personalized recommendation technology is widely used in e-commerce, while the algorithm is being replicated to the adaptive learning system, divided into content-based filtering recommendation algorithms and collaborative filtering recommendation algorithms. Finally, in addition to recommendation system itself, in order to achieve the recommendation, but also recommended available content set. For collaborative filtering recommendation system, the content set or even just need to provide ID enough; but for content-based recommendation system, because the content is often the need for feature extraction and indexing, we will need to provide more domain knowledge and content attributes. In order to give learners a small amount of labor as much as possible, the system can accurately recommend to learners learning content interested in, this study suggests that the content sets can be stored in some emotional imagery information of the learners for system quickly capturing the learner's preferences. So in the personalized content recommended strategy, reference some of the design steps of the kansei engineering. Organizational Form of Content At present, many adaptive learning systems have drawn on the knowledge-point as the center of Intelligent Tutoring System (ITS) to conduct research. At the same time, knowledge representation becomes the difficulty of technology. In addition, some organized strategy of the teaching content should also be applied to the adaptive learning system design. First of all, follow the basic principles of teaching content organization such as the combination of knowledge sequence and cognitive order, the association between knowledge, organization optimization. Secondly, the organizational form of content should be: offering different relatively independent themes; under a theme setting different a number of topics of diverse views; using examples of varying degrees or multiple perspectives (Fig.3).

Fig. 3. Organizational Form of Content op

212

z

z

L. Bian and Y. Xie

intuitive learning materials-based, to enhance memory of learner; experience-based, to promote knowledge application. To adjust desire level. First, the system understands the needs through link autonomy clicked by learner; secondly, describes information about learner’s preparation; finally, shows the causes of recommendation corresponded with the real level and the future benefits of learning, encouraging learner’s confidence. To bring up learning perseverance. The creation of adequate difficult situations for learners’ different circumstances. The system also gives the necessary guidance and encouragement, to guide in the learning process for self-examination, self-affirmation, self-monitoring and self-encouragement.

5 Mode of Study to Choose 5.1 Theoretical Analysis At present, the learning methods according to different criteria are divided into various categories. This study suggests that adaptive learning system can provide the three kinds: accept-learning, inquiry-learning and collaborative-learning, for learners according to their own needs to choose the mode to learn new content. Accept- learning is that learner masters knowledge by material presented which is from Ausubel’s meaningful reception learning. Inquiry-learning is that the learner independently identifies problems, experiment, operation, investigation, information search and processing and exchange exploration in order to acquire knowledge, skills, emotions and attitude development. Collaborative-learning is that by using of computer networks and multimedia technologies, a number of learners interact with each other for the same learning content and cooperate in order to achieve a deeper understanding of the teaching content. There are four basic strategies which are competition, collaboration, partnerships and role playing. 5.2 Design Model of Supporting Accept-Learning z Advance organizers presented. Systems need to clarify the course objectives, show and explain advance organizers, to arouse learner’s prior knowledge and experience. Among them, advance organizers is not refresher of previous content or the recall of contents which have be learned, nor is it the memory of learners previous experience. Its essence is a kind of e concept which is more abstract and general than learning materials, is also clear and detailed instructions of the basic characteristics of concept or principle. Advance organizers are divided into three categories: upper organizers, the next bit organizers, tie organizers. [7] z Presentation of a new learning content and organizational structure. This part mainly completes by dynamic organization and presentation of the contents of adaptive strategy. z Migration and application of new knowledge learned. System asks questions, leads learner to positive think and use new knowledge, the final uses post-learning diagnosis to determine whether outcomes or learning goals been achieved.

Research on the Adaptive Strategy of Adaptive Learning System

213

Model of Supporting Inquiry- Learning z The creation of situation. System creates situation to stimulate motivation through a variety of ways such as play the video and audio related to the current learning topics, reviewing old knowledge related to the current knowledge, teaching games, teaching Flash animation, virtual reality rendering tools, showing homework been completed, an activity the introduction of the subject. z Inspiration. Systems need to giving advice and guidance such as how to solve the problem, how to complete the current task, using what kind of learning tools or learning resources, how to use tools and resources to explore and so on. z Self-exploration. Cognitive tools provided for learner to solve the problem can be divided into four categories. Tools of optimize the effectiveness such as calculator, search engine; tools of learning content management such as Notepad, online bookmark; tools of learning information communication such as E-mail, BBS; tools of learning disabilities elimination such as the course FAQ, expert systems, on-line Q & A Center. z Summed up to improve. System gives summary of teaching knowledge points, comments learning, produces other problems or situations of migration. Model of Supporting Collaborative- Learning System generally asks a question. Instructor identifies learning tasks and collaborative goals for the problem, and guidance, supervision and coordination of collaborative learning process. It emphasize that the role (instructor or learner) can be interchangeable in the learning process. z System provides all relevant information to help solve the problem. Learning contents and collaborative learning outcomes are stored in the learning resource which can be visited at any time according to needs. z Evaluation of instructor and group member is recorded into cooperative groups / members. And the result can be seen as reference of learning diagnosis. z

Application of the "Mutual Adaptation" Idea Analysis found that learners will have non-intellectual factors such as relationship between student and student, reading habits, note method which affect their adaptation to mode of study to choose. Therefore, the system imposes these areas assistance strategy through “adjust strategy to encourage learners to choose collaborative learning”, “reading tools embedded”, “dual-screen of left and right”.

6 Conclusion In this paper firstly used literature reading method to identify the development of adaptive learning system and several adaptive techniques, also founded mutual adaptation relationship among adaptive learning, and put forward the "Mutual Adaptation" idea applied in the system design. Simultaneously, the design of adaptive strategy which is learning diagnosis, dynamic organization and presentation of the contents, mode of study to choose, had given to add by the application of the idea. In addition author hopes that these non-intellectual factors which affect the learner to adapt to the system could be extracted into a part of the student model. And how to quantify these factors has become a next effort of adaptive learning research.

Research on the Adaptive Strategy of Adaptive Learning System

211

Presentation Forms of Different Learning Styles In accordance with a detailed description of three kinds of learning styles (Mary.Kabo and Dr. Dunn): visual, auditory, kinaesthetic. So, the system should be based on learner's learning style to select the most suitable presentation forms (Table 2). Table 2. Presentation Forms of Different Learning Styles

Visual Orientation-Type

z z z z z

z

Auditory Orientation-Type

z z

Kinesthetic Orientation-Type

z z

Describe the content of teaching with determinate language which is concise, clear and suitable for student’s reading level Pages of text formatting consistent and to provide the necessary tips in order to facilitate reading A reasonable content’s length to avoid too fragmented or cause cognitive overload Use of next page way between old and new contents, in order to facilitate reading, to prevent the confusion of old and new contents Using a graphical image of the text to make the necessary explanations, animation examples of teaching content to make the necessary demonstration ,for helping learner remember or understand Using color, the color of mainly content is a dark green which should be the most vulnerable for the human eye to accept; a little light green is used for highlight the text; red is used for keyword Audio sounds clear, standard, smooth, there is no ambiguity on the voice Explain synchronization of electronic speech and audio. The electronic speech content is simple and refining Provide substantial opportunities for hands-on exercises and operations Development WordPad so that students take notes and record Essay

Application of the "Mutual Adaptation" Idea Analysis found that learners will have non-intellectual factors such as learning motivation, desire level, enthusiasm for learning, learning perseverance which affect their adaptation to dynamic organization and presentation of the contents. Therefore, the system imposes these areas assistance strategy through the following design. z

To stimulate learning motivation and enthusiasm for learning. The ways are multimedia-based, to stimulate the curiosity of learner and maintain attention;

214

L. Bian and Y. Xie

References 1. Khan, B.: Managing E-Learning Strategies: Design, Delivery, Implementation and Evaluation. Idea Group Inc., Hershey (2001) 2. Li, K.: Network Environment CBE Research a New Topic. Zhejiang Normal University lectures, R. Jinhua (2003) 3. Conlan, O.: The Multi-Model, Metadata Driven Approach to Personalised eLearning Services. Knowledge and Data Engineering Group, Department of Computer Science, Dublin 4. Bian, L., Xie, Y.: Research on Mutual Adaptation problem of Adaptive Learning System. J. China Educational Technology 3, 9–12 (2009) 5. Chen, Y., Parsons, R.D.: Educational Psychology: Practitioners - the Road of Researchers, vol. 145. Shanghai People’s Publishing House, Shanghai (2007) 6. Jing, M.: Middle school math teacher’s teaching content knowledge development strategy study. China East Normal University, D. Shanghai (2006) 7. He, K., Wu, J.: Information Technology and Curriculum Integration, pp. 147–153. Higher Education publishing house, Beijing (2007)

Research on ducational Software Defect Prediction Model ased on SVM Guang-jie Liu 1 , and Wen-yong Wang 2 1

Department of Computer Science and Technology, Changchun Normal University, Changchun, 130032, China 2 Software College of Northeast Normal University, Changchun,130024,China liuguang [email protected], [email protected]

Abstract. We must pay attention and find defects, defects through the prediction to quantify the quality management and quality in order to achieve this goal, requires an estimate of the various defect detection process. Software defects are the departure of software are products¶ anticipative function. This paper collecting the data of the software defects, then, using the SVM model the predictive values are gained analyzing the predictive results, software are organizations can improve software control measure software process and allocate testing resources effectively. Keywords: Software defect; SVM; Software lifecycle; Educational software.

1

Introduction

The current product quality is no emphasis on educational software, because the educational software than industrial control software requirements less stringent, so quality control is often neglected. With the look in-depth education of information technology applications, has been fully applied to the teaching of information technology and management of information technology, by which software quality requirements are increasing. For example the organization provinces, municipalities and academic proficiency test to participate in thousands of students, but these tests relate to the future and destiny of the students, once the error, will cause a great impact, so the need to improve the quality of educational software to limit the mistakes, so that error rate minimized. There are many quality features software to quantify the quality management are two key aspects: setting quantitative quality goals and quantitative management of software development process in order to achieve quality goals [7]. Quality management is often around the defects, while software development is a high degree of human activity - so easy to make mistakes. Therefore, in practice, the use of the density of defects presented in the submission of the software the number of defects per unit size as the definition of quality[1].For the high-quality software, shortcomings of the final product should be as little. Quantitative control software quality and software reliability model is to work together. Most software reliability models using unreliable data to estimate the reliability of software. The reliability of X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 215–222, 2010. © Springer-Verlag Berlin Heidelberg 2010

216

G.-j. Liu and W.-y. Wang

these models are able to identify whether it is acceptable, or whether the need for more testing. On the whole, such a model would help to assess the reliability of software products, but the software quality management have only limited value. Good quality management methods should have the option is limited to an early rather than late in the project to provide warning signs. Through early warning is able to intervene in a timely manner. In order to achieve this goal, there must be various stages of the project predicted values of some parameters, so the project can control these parameters in order to ensure the final product has the expected quality [8,9]. In the software life cycle are likely to introduce defects at any stage. That is, the user needs into a software process to meet those needs, in the implementation of any conversion activity may be injected into defects. These phases include demand for standardized, the outline design, detailed design and coding. The project process, including many of the activities of identifying defects, and these activities can remove defects. Despite the recognition and removal are two distinct activities, but we will use clear to jointly refer to them. The longer incubation period of defects, the cost of the more clear it. Therefore, any mature process will be injected in each stage of defects include quality control activities (as shown in Figure 1). Defect removal activities include needs assessment, design review, code review, unit test (UT), integration testing (IT), System Test (ST), and acceptance testing (AT). Development Process Defect injection

Requirements Analysis

R

Design

R

Coding

R

UT

IT/ST

AT

R ± Remove UT - Unit testing IT - Integration Testing ST - System Testing AT - Acceptance Test Fig. 1. Defect injection and removal

Defect removal

Research on a Educational Software Defect Prediction Model Based on SVM

217

Support Vector Machine (SVM) is based on statistical learning theory, developed a completely new machine learning algorithm, which is based on statistical learning theory of structural risk minimization principle, maximize the class interval of ideas and kernel-based methods together, showing a good generalization ability. It can solve the small sample, nonlinear and high-dimensional pattern recognition practical problems, and overcome the learning method of neural network is difficult to determine the network structure, slow convergence, local minima, over-learning and less learning, and training requires large amounts of data samples, and inadequate.

2

Support Vector Machine Model Description

For linear classification, support vector machine The main idea is through the known training samples for training, construction discriminant function, that is to find an optimal hyper-plane, so that the largest class interval. For the non-linear case, SVM through the non-linear transformation, a sample of the original space is mapped to a high-dimensional feature space, in this high-dimensional feature space to find an optimal hyper-plane, so as to solve the original space for Central Africa non-linear sub-problems [2]. Input-based non-linear discrete model as follows: x(n) [ x(n), x(n  1),, x(n  N  1)], output:˖Y (n) xÖ(n  1), Suppose a total of l input-output pairs, then the SVM in the following estimation function˖ y

g ( x)

Z T I ( x)  b

˄1˅

Where I ( x ) from the input space to high-dimensional feature space of the non-linear mapping; it is an adjustable weight vector; b is the bias; coefficient and b by minimizing the estimated cost functional. Make [3] Rsvm (c)

c

1 n 1 ¦ LH (Yi , yi )  2 Z n i 1

­ 0  Y  y d H ® Y y  H  Y  y  ¯

LH (Y , y )

Form˄2˅In the first part of the 1

2

2

c

1 n

n

¦ LH (Y

i

i

, yi )

˄2˅

H

˄3˅

is the experience of risk,

1

the second part of the 2 Z is a regular part of. C called the penalty function, determines the experience of risk and regularization of the balance between the parts, C larger, which means more weight on the wrong punishment. In general, as C increases testing accuracy will be getting higher and higher, but when you reach a certain value. However, further increase will make the error rate increases. There is no one single best way to determine the C values, the general approach is to test, through continuous testing to get satisfactory results. To solve the optimal and b, the need to introduce slack variables to be constrained optimization˖

218

G.-j. Liu and W.-y. Wang

min

n 1 || Z || 2  ¦ ([ i  [ i* ) 2 i 1

s.t.Yi  (Z T I ( xi )  b) d H  [ i (Z T I ( xi )  b)  Yi d H  [ i*

[ i , [ i* t 0, i

(4)

1,2,  N

Lagrange multiplier method, the original problem that the optimal separating surface is converted to the dual problems of solving the problem-solving, then the formula (4) The dual optimization problem written in matrix form as˖

max E

T

a

a

1 T a Pa 2

n

s.t.¦ (ai  ai* )

0

˄5˅

i 1

0 d ai , ai* d C Which˖ a T

ET

[a1 ,, an , a1* ,an* ],

[H  y1 ,H  yn , H  y1 ,H  yn ] ˈ P

n-order symmetric matrix and the Qij Lagrange multiplierˈ vector [4] [5].

a

i

ª Q « Q ¬

 Qº Q »¼

Q is an ˈ known as the

I ( xi ) T I ( xi ) . ai , ai*



 ai* z 0 corresponds to the sample is called support

Support Vector Machines commonly used kernel functions are polynomial kernel function, radial basis kernel function, Sigmoid function. In this paper, radial

basis function that is K ( xi , x j )

exp(

xi  x j 2V

2

2

) ˈreceived support vector

machine is a kind of radial basis function classifier, in which the nuclear function. n

Solving the above optimization problem, available

Z

¦ (a i 1

i

 ai* )I ( xi ) ˈ

accordance with the KKT theorem [6], derivation of equations ˖

­H  y i  g ( xi ) ® ¯H  y i  g ( xi )

0 ai  (0, c) 0  ai*  (0, c)

˄6˅

Research on a Educational Software Defect Prediction Model Based on SVM

219

A bias value b can be solved. Is forecast to be the decision-making function is

g ( x)

SVx

¦ (a

i

a *i ) K ( xi , x)  b

˄7˅

i 1

3

Software Defect Prediction xperiments

3.1

Data Acquisition

Software defect subject to many factors, processes, and technology, often referred to as the quality of triangle. For example, design methods, development of personnel quality, management level , development environment , testing strategy, the maintenance activities. Developed with input costs, the provisions of the relevant delivery period. In this paper, to consider the various stages in the software life cycle, a measure of yuan [1]. Table 1ˊDefect Detection Phase

Detection stage Requirements Specification Project Configuration Management Outline Design Detailed Design Unit Testing Code review Separate unit Integration Testing System Testing Acceptance Test

Defect Examples Incorrect or too bold assumption; process map does not recommend Chu Incorrect amount of work or progress estimation; not true of human-use plans Unreasonable configuration management agencies; access control problems No integrity; no maintenance tracking Logical problems; The wrong test; not determine the correct test data; use case is not sufficient Related issues with the coding standard, redundant code, logic or style And logic, data processing or input / output-related issues Test is not sufficient; test case does not involve the interaction between modules All types of defects And function, the external interface, input / output or performance-related issues

This collection of educational software development project in the 11 practical software measurement data, a total of 412 modules. Of which the first eight software as training samples, after the three software as test samples.

220

3.2

G.-j. Liu and W.-y. Wang

Forecast Analysis

This simulation of all the procedures were realized under Matlab 6.5 environment. Take the time delay T = 1, take n = 12, through cross-validation selecting the optimal penalty parameter C = 120, = 0.0001. Radial basis function parameters = 1, and (C,) = (120,1) when the training model; previous 11 samples as training set, the latter three samples as a test set. Get SVM software defects such as Table 2. Table 2. SVM predicted results compared with the actual value

Requirement specification + HLD + detailed design of the assessment (actual value) Code review + unit testing (actual value) Integration testing + system testing (actual value) Acceptance testing (actual value) Requirement specification + HLD + detailed design of the assessment (prediction value) Code review + unit testing (predicted value) Integration Testing + System Test (predicted value) Acceptance Test (predicted value)

Test samples 1 45

Test samples 2 11

Test samples 3 13

148

62

81

64

15

16

27

6

8

45

12

14

150

61.5

84

62

14

14

28

5.7

7.5

(Obtained MSE = 0.894. MSE is the characterization of the test samples were the actual values and predicted values of the average deviation from standard.) 3.3

Defect Analysis and Prevention

Concerned about the defect analysis helps to understand the characteristics of defects and defect prevention in the future. Therefore, SVM can be used to carry out institutional-level and project-level defect analysis. Agency-level defect analysis can improve the body within the checklist, process or training. The idea behind this analysis is found flaws in the project to learn, and future projects to improve organization-wide process. On the other hand, the shortcomings of the project-level target is found in the project from the date the defects in learning and the rest of the project to prevent defects. We consider the project-level focus on defect analysis, according to defect classification rather than that of stage to understand the characteristics of defects [10, 11].

Research on a Educational Software Defect Prediction Model Based on SVM

221

Table 3. Defect Type

Defect Type Logic Standard Redundant code User Interface Performance Reusability Design Memory management flaw Document defective Consistency Traceability Portability

Examples of defects The algorithm used by inadequate / incorrect; wrong conditions, test cases or design documents Coding / documentation standards, such as indentation, alignment, layout, component applications. In many programs or the same program using the same program segment The designated function keys do not work; not correct menu guide Processing speed is very slow; because the file size Ershi systems collapse; memory problems There is no ability to reuse code And design issues related to the specific Information dump, an array of cross-border, illegal function calls, the system hangs or memory leaks, etc. In assessing the deficiencies found in a document, such as project planning, configuration management plans or specifications By the same order of update or delete records of the entire system Program source code to the specifications traceability Code is not platform-independent

Defect Type Distribution User interface

25 20 Logic

Coding Standard

15

10 Boundary portability performance design matters other 5 0

Fig. 2. Defect Type Distribution

On the project in terms of defect prevention activities on the stage of the analysis of defects identified, and at this stage is usually found in the remaining part of the defects have a greater association. For example, in building a stage, when the number of modules have been built, and there is sufficient data for defect analysis, defect analysis can be carried out. Facts have proved that this analysis results in the remaining part of the build phase is very useful. At a minimum, recommended the implementation of the seriousness and type of defect analysis. If this analysis shows that an overwhelming majority a defect types, then the process can be improved to prevent such defects.

222

4

G.-j. Liu and W.-y. Wang

Conclusion

The inherent properties of the software makes the software system defects in the software development process, the inevitable "by-product." At present, many software defect tracking software can be found in tests on full record of the defects, which for statistical and evaluation of software defects bring great convenience. But how reasonable to use these historical data to predict software defects and needs further study. In this paper, the number of software defects historical statistical value, based on the software life cycle, based on the use of support vector machine method, the practical engineering of software defects were predicted through the calculation, testing and verification, found that predicted a good result. Shows that this method can guide software organizations to better grasp the software quality to help the institution-building CMM, in order to improve their software development process capability .

References 1. Jalote, P.: CMM in Practice-Processes for Executing Software Projects at Infosys, pp. 94–95, 172–175. Electronic Industry Press, Beijing (2002) 2. Zhi, S.: Knowledge Discovery, pp. 7–8, 213–215. Tsinghua University Press, Beijing (2002) 3. Yang, J.-E., Wei, C.: China. Based on support vector machine safety input demand forecasting research. Coal Economic Research 9, 84–85 (2009) 4. Lin, Z.X., Wei, L., Bo, X.C.: The new discrete time-varying delay system stability condition. Journal of Northeast Normal University 3, 31–36 (2008) 5. Bennett, K.P., Campbell, C.: Support Vector Machines: Hype or Hallelujah? SIGKDD Explorations 2(2), 1–6 (2000) 6. Barabino, N., Pallavicini, M., Petrolini, A.: Support vector machines vs multi-layer perceptrons in particle identification. In: Proceedings of the European Sympostium on Artifical Neural Networks’99, pp. 257–262. D-Facto Press, Belgium (1999) 7. International Standards Organization, Information Technology-Software Product Evaluation-Quality Characteristics and Guidelines for Their Use. ISO/IEC IS9126,Geneva (1991) 8. Kan, S.H.: Metrics and Models in Software Quality Engineering. Addison-Wesley, Reading (1995) 9. Musa, J.D., Iannino, A., Okumoto, K.: Software Reliability-Measurement, Predication, Application. McGraw Hill, New York (1987) 10. Grady, R., Caswell, D.: Software Metrics: Establishing a Company-wide Program. Prentice Hall, Englewood Cliffs (1987) 11. Grady, R.: Practical Software Metrics for Project Management and Process Improvement. Prentice Hall PTR, Englewood Cliffs (1992)

Webgame Based Collaborative Learning Design: A Case Study Jie Jian1,2, Yueguang Xie1,2,3, Wenhe Tang1, and Chunhui Wang1 1

School of Software in NorthEast Normal University, ChangChun China, 130117 2 Engineering & Research Center of E-learning, ChangChun China, 130017 3 E-learning Laboratory of Jilin Province, Changchun, Jilin, 130024 [email protected], [email protected], [email protected], [email protected]

Abstract. The purpose of this paper is to design a schema of webgame-based collaborative learning. Using the most popular webgame “Happy Farm” as a case study, this paper reviews and analyses the characteristics of the game and usability of the interface through learning theories such as behaviorism, cognitivism and constructivism and interface design theories by Shneiderman and laws of gestalt psychology. It indicates that webgames can be reused to be good learning resources if designers integrate reasonable topics and tasks. They can provide a collaborative learning environment and are good for developing learners’ higher order thinking skills. Usability principles used in this case study should be provided by the teachers to choose webgames to ensure more standardized and rigorous learning environment. This learning design provides teachers with insights into the use of learning theories and usability principles for choosing webgames for designing learning activities. It also provides a gauge of current usability level of commercially developed webgames as usage of such webgames continue to proliferate. Keywords: Webgames; Collaborative learning; Higher order thinking skills; Case study.

1 Introduction Learning is no longer confined to textbooks and classrooms. They now roam the virtual world for entertainment and educational purposes. With this comes the era of “Edutainment”, that is, to be sufficiently entertained while achieving learning goals. In order to appeal to school-going learners whose experience with computers began early, more attention should be paid to the specifics on how to design for them. The importance of ensuring authentic learning through application of learning theories to instructional design cannot be downplayed. 1.1 Background Information and Contributions According to the “Competition in the 21th Century” report, skills of being and developing in the knowledge age are mainly ten aspects, including communication skills, X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 223–234, 2010. © Springer-Verlag Berlin Heidelberg 2010

224

J. Jian et al.

innovation and creativity, team collaboration and organizational capacity, information management capacity, information technology literacy, visual literacy, problem-solving ability, decision-making ability, knowledge development and management capabilities and business intelligence. There are American scholars (Trilling, B. & P. Hood, 1999) who advance seven basic skills acquired by the information era based on integrate researches. The seven skills are: critical thinking-and-doing, creativity, collaboration, cross-cultural understanding, communication, computing, career and learning self-reliance. So collaborative learning is important, especially in the development of cognition construction. In collaborative learning, learners are in charge of their study. It indicates that 80% learners can do self-criticism, while 20% learners can do that in non-collaborative learning. After that, collaborative learners often play several roles, such as manager, inquiry man, cognitive apprenticeship, teacher and knowledge producer, and that is good for obtaining social experience. Collaborative learning is an effective way of developing learners’ high order thinking. Webgames are so popular nowadays that people from old to young, from boys to girls, are involved. This naturally forms a wide communication environment. As games are usually full of fun, they can cultivate people’s interest and motivation. With the networks being more and more convenient, people can access to webgames anytime and anywhere. It always takes a lot of time for teachers and educational experts to make digital resources for specific study. Some excellent webgames can be reused to be good learning resources if designers integrate reasonable topics and tasks. Webgames can provide a collaborative learning environment. In this paper, we will advance a learning design of webgame based collaborative learning. And we’ll introduce a very popular webgame in China now, called “Happy Farm” as a case study. 1.2 Terms Defined Webgame is one kind of the online multiplayer games based on Web browser. It is roughly divided into two types. One is using interpreted languages such as PHP / ASP / Perl / JAVA to build virtual communities; the other is using Flash technology to make the game. Webgame is on B / S mode which has the advantage of non-client. Users do not need to download any client or plug-in, but to open the browser directly and visit the website of the game. Since the emergence of Ajax techniques, Webgame becomes easier to be developed and the browser-based interactive game becomes entirely achievable. With the development of broadband, Webgames will be important complement of webgames. Collaborative learning is a situation in which two or more people learn or attempt to learn something together (Dillenbourg P., 1999). More specifically, collaborative learning is based on the model that knowledge can be created within a population where members actively interact by sharing experiences and take on asymmetry roles (Mitnik et al., 2009). Put differently, collaborative learning refers to methodologies and environments in which learners engage in a common task where each individual depends on and is accountable to each other. Collaborative learning is heavily rooted in Vygotsky’s views that there exists an inherent social nature of learning which is shown through his theory of zone of proximal development (Lee et al., 2000). Thus, collaborative learning is commonly illustrated when groups of students work together to search

Webgame Based Collaborative Learning Design: A Case Study

225

for understanding, meaning, or solutions or to create an artifact or product of their learning. Collaborative learning activities can include collaborative writing, group projects, joint problem solving, debates, study teams, and other activities. Higher order thinking and learning can be broadly defined as an expansion of the mind to meet new challenges. Hopson, Simms, and Knezek (2001) define higher order thinking skills as the cognitive skills that permit learners to perform at the analysis, synthesis, and evaluation levels of Bloom’s Taxonomy. Lewis and Smith (1993) purport, “Higher order thinking occurs when a person takes new information and information stored in memory and interrelates and/or rearranges and extends this information to achieve a purpose or find possible answers in perplexing situations” (p.136). This definition implies that higher order thinking skills include critical thinking as well as problem solving, creative thinking, and decision making.

2 Gaming Environment "Happy Farm" is a social game which is developed by Five Minutes Co. (Shanghai, China). Users can play as farmers on their own farms planting a variety of vegetables and fruits. At present, Everyone Network (renren.com) which is the largest and most influential SNS site in China has implanted this game at the end of 2008; QQ (qq.com) which is the earliest and largest Internet instant messaging software developer in China has implanted it on May 22, 2009; BIDU (baidu.com) which is the world’s largest Chinese search engine has implanted it on August 27, 2009. Game players assume the role of a farm operator to do seeds purchasing, cultivation, watering, fertilizing, pesticides spraying, harvesting and finally sell them to the market. The crop will go through different phases from planting to maturity. During each phase, there may be drought, pests, weeds and other disasters. So farmers should take care of the crop. When vegetables and fruits are mature, they can be harvested into the warehouse. Especially, that is the time for players to steal vegetables and fruits from other players’ farms. Crop in the warehouse can be sold for gold which can be used to buy seeds, fertilizer, and farm decorations, as well as for expansion of farm land. 2.1 Characteristics The Webgame "Happy Farm" emphasizes mutual interaction which means the more friends the more fun. Players can take care of the crop in their own farms or friends’ farms on the Webgame every day, such as watering crops, killing pests and clearing weeds. The game can mobilize the enthusiasm of players, and encourage them to interact with more friends. The simulation of crop growth process brings fun of cultivation to players. It supports the following characteristics: 2.1.1 Emphasize Interaction Interactions are emphasized among players. In addition to watering crops, killing pests and clearing weeds for friends to enjoy the fun, players can also occasionally trouble friends with putting pest, growing weeds and stealing friends’ ripe fruits. Interactions are also emphasized between players and the game as the interactive interface-design.

226

J. Jian et al.

2.1.2 Easy to Engage The game simulates real farming procedure so that players can easily understand the basic operating process. As so many people participate in it, it makes players confident of control the game and easy to engage. 2.1.3 Play Anytime and Anywhere The game can be accessed whenever you connect to the Internet and log on the webpage. And Players can use PCs, Mobile Phones and Mobile Internet Devices to get the game. 2.1.4 Constructivist Approach Players are prerequisite with computer knowledge such as keyboarding skills, usage of mouse and web-browsing. When considered as learning environment, the Webgame can be reused as a type of e-learning system which falls under the category of “School, Home Entertainment Applications” whereby ease of learning the program, low rate of errors and subjective satisfaction are important as the use is frequently discretionary and competition is fierce (Ragozar, 2005). 2.2 Interface Usability Raskin (2000) defined interface as “the way that you accomplish tasks with a product – what you do and how it responds – that's the interface”. In order for students to learn, an interface cannot be poorly designed to have them feeling lost, confused or frustrated which will pose difficulty for effective learning and information retention. Thus, usability of the interface needs to be considered. Using theories from Schneiderman's “eight golden rules of interface design” (Schneiderman, 1998), this paper documents the use of usability principles such as feedback, consistency, recognition and visibility for evaluation of interface of the game “Happy Farm”. 2.2.1 Offer Informative Feedback Children acquire the knowledge of cause and effect since young so they would expect to receive responses from their actions. As such, input devices should have direct mappings to the actions on the screen (Chiasson and Gutwin, 2005). A brief animated action should also occur when children mouse over and click an icon. This will assist them to know that a function has been activated which is associated with the icon (Dunham and Sindhvad, 2005). Visibility of system status when the computer is busy processing requests should be displayed so that children anticipate the next step. This can be achieved through simple on-screen icons or audio feedback to facilitate understanding. Conversely, if the system has been expecting input from users for an extended period of time, it should reflect with feedback such as humming or toe-tapping (Hanna et al., 1999). In the case of the “Happy Farm” game, when learners mouse over the button on the menu bar, a brief animated action (button size changes or color changes) takes place so they are aware that an action is associated with the button. When the button is clicked, there is audio feedback provided through the clicking sound. During the process of waiting, the system also includes messages such as “loading” plus a progressing bar

Webgame Based Collaborative Learning Design: A Case Study

227

Fig. 1. Loading message plus progressing bar

showing the status. This helps to reduce learners’ anxiety of thinking that the program is not functioning properly (Figure 1). However, an area of improvement would be for the system to prompt learners for input after an extended period of non-activity. 2.2.2 Consistency A fundamental part of mental model would be button identification. There should not be any similar design characteristics if the on-screen graphics are not meant to be buttons. Otherwise, learners may think there is a new button with new functions if its location or appearance has been changed. A three-dimensional button could be used to help them identify that an action can be taken. Other ways of indicating that an object is clickable may be through a change in the mouse cursor or by using roll-over effect. However, it must be done consistently for every clickable object and should have the same visual effect (Kruse, n.d). All buttons in the game have the same three-dimensional effect when moused over. There is consistency in the shape, location as well as images of the buttons. The basic operation buttons are consistently placed on the right-hand side of every page which will help learners to click more easily. It also reduces the load on memory as learners do not have to search for it (Figure 2).

Fig. 2. Operation buttons are consistently placed on the right-hand side

228

J. Jian et al.

2.2.3 Aesthetics and Design of the Portal Visibility Children often click visible on-screen features in order to see its effect (Sullivan et al., 2000). As such, buttons need to be placed where the user's eye can easily find them with clear symbols or labels given. Given the Z-pattern of reading (from left to right, and from top to bottom), children hardly scroll pages and mainly interact with the information visible above the fold (Gilutz and Nielsen, 2000). In addition, buttons should represent familiar things to children, convey their purpose easily and be fairly large to accommodate maturing hand-eye coordination (Hanna et al., 1998). For the screen layout of the “Happy Farm”, the menu for navigation buttons is always positioned on the top of the interface (Figure 3 & Figure 4). It is appropriate since we understand their Z-pattern of reading whereby the eyes will browse the screen from top to bottom. Another positive point is that scrolling of pages is rarely required. There are also visible actions and feedback mapped to buttons and icons activated by learners through animation. The buttons are reasonably sized and in simple recognizable shapes with text that conveys its purpose.

Fig. 3. Navigation buttons are always positioned on the top of the interface

Fig. 4. An example of navigation information

Webgame Based Collaborative Learning Design: A Case Study

229

2.2.4 Gestalt Psychology: Law of Organization As propagated by Chandler (1997), buttons should be grouped together based on their function (principle of proximity). Screen objects or menu items need to be placed together in a logical order to match the structure of the tasks (principle of continuity). The structure of the functional buttons in the game is logical and universal as it mirrors the operation habits. This familiar context and structure provides scaffolding to learners and shorten the time for transition from novices to experts. Buttons are classified into three groups, including navigation buttons (on the top), basic operation buttons (on the left) and common management buttons (on the bottom) (Figure 5).

Fig. 5. management buttons are on the bottom

2.2.5 Usage of Interface Metaphors Borrowed analogies from the real world help our learners to establish their expectations and predict what is likely to happen (Helander et al., 1997). For example, the “Happy Farm” uses the familiar story of working in the farm. As learners associate positive feelings with this metaphor, it will help them to better relate with learning on this platform of game. This contextual knowledge is familiar and links to learners’ existing schema and hence reduces the load on short term memory. This is particularly powerful for our learners who fall in the concrete operations stage. It will help them understand how things work. 2.2.6 Structuring Information Using Hierarchies Due to short-term memory constraints, most users begin to lose track of their location or the relationship of the on-screen content to the overall lesson when a menu contains more than three layers or paths. It will help learners to remember which menus contain certain items if they can be split logically into sub-menus (Kruse, n.d). 2.2.7 Engagement The study by Gilutz and Nielsen (2000) observed that “animation and sound effects often help in creating a good impression of a site and encourage children to stay with the site”. Since children are much more attracted to animation than adults, they will

230

J. Jian et al.

click an area on the screen just to continue the animation or sound whenever there is a resulting effect (Bernard, 2003). Within the Happy Farm, learners are immediately brought into a world of animation and sound as soon as they open the game page. This should be able to engage the learners throughout the whole session especially with busy farm work.

3 Collaborative Learning Design Collaborative scripts structure collaborative learning by creating roles and mediating interactions while allowing for flexibility in dialogue and activities. Collaborative scripts are used in nearly all cases of collaborative learning, some of which are more suited for face-to-face collaborative learning—usually, more flexible—and others for computer-supported collaborative learning—typically, more constraining (Dillenbourg et al., 2007; Kollar et al., 2006). Additionally, there are two broad types of scripts: macro-scripts and micro-scripts. Macro-scripts aim at creating situations within which desired interactions will occur. Micro-scripts emphasize activities of individual learners (Dillenbourg et al., 2007). 3.1 Conceptual Components of Scripts The conceptual components of collaborative learning scripts are: (1) objectives, which help participants work together to engage in efficient collaboration processes to reach specific objectives; (2) activities, which include summarizing, questioning, giving an argument, state a claim; (3) sequencing, which explain the expectations of the participants by specifying which activities should be performed and in what order; (4) roles distribution, which will assume throughout the activity to encourage participants to adopt and consider multiple perspectives; (5) type of representation, including textual, graphical, or oral representations of explicit instructions, which are presented to the participants (Kollar et al., 2006). 3.2 Objectives Considering the Happy Farm as a collaborative learning environment, it can be constructed to be a practical field of plan, manage, share, and self-control. Learners are encouraged to make a plan of managing their farms, independently or collaboratively, in order to achieve the target set by the teacher, which includes certain quantity of coins, experience value, capability level and kinds of crop in the warehouse within 7 weeks. Of course, as it’s a collaborative learning activity, teachers should conduct learners to find that teamwork always does a better job than individual. Through the process of experiencing the fun of planting, watering, stealing, harvesting and selling, leaners are trained to share with others, to care for crop, to think of selling and purchasing. In addition, as the game is so easy to be addicted that it’s an important challenge for learners to manage their time and spirit on the game.

Webgame Based Collaborative Learning Design: A Case Study

231

3.3 Activities The learning activities are designed for learners to be happy farmers. Firstly, as a farmer, learners have to manage their farms in reason. In this design, the managing target is to obtain coins and experience as much as possible, and to store agricultural products as more as possible. Secondly, in the real world, the learners are not permitted to manage their farm in the virtual world with too much time. They need to arrange their time rationally to do common study work and play games. The design will help learners to control their willpower and prevent them from indulging in games. Thirdly, the feature of the game “happy farm” is stealing vegetables among other game players to experience the special fun. Our learning design means to make learners enjoy themselves in this game. At last, there will be a discussion for students to examine their behavior during the whole process and rethink the significance of the collaborative learning. 3.3.1 Submit a Project Plan Submit a project plan to achieve the farming target. In the plan, students need to write down their own understandings of the game’s rule, as well as their business plan, including list of the seeds purchasing, order of the cultivation, projections of the harvest, plans of the agricultural products selling, and the theft scheme. 3.3.2 Develop a Schedule of Gaming Develop a schedule of games. Students need to list the time and place when and where they play the game. And they are encouraged to think out ways of monitoring their own practices to prevent themselves from addicting to the game as much as possible. 3.3.3 Enjoy the Game When enjoy the game, students may add more friends to their list of players. Through the interactive communication with so many people, students will enjoy the fun of social communication. When enjoy the game, students may experience the process of planting, managing, harvesting and stealing. 3.3.4 Tell Us Your Feeling Tell the teacher and all your classmates your feeling about this collaborative learning activity by answering the following questions (not limited to): ·Have you experienced the fun of stealing vegetables? ·To prevent yourself from being stolen, have you adopted any strategy? ·Who do you think is the most astute farmer? Why? ·Do you think your project planning is a success? Or what are the deficiencies? Why is there such a shortage? ·Have you played the game in accordance with your timetable? ·Do you often indulge in the game? ·Do those ways of monitoring work? ·Do you like this game? What have you learned from it? The learning activities can be conducted not only in the school's computer room when having information technology class, or comprehensive practice activity class, but also in learner's home as well as any place where can be connected to the Internet.

232

J. Jian et al.

3.4 Sequencing During the 7 weeks, before starting the game, preparations includs registrating the site, adding game applications, learning game rules and training basic operations of the game interface. These are mainly conducted by the teacher and will take one week time. After this, learners need to make plans of farm managing and game scheduling with patners. This still will take two weeks time. During the next three weeks, students can enjoy the game according to the managing plan and game scheduling. In the last week of this activity, the teacher will hold on a discussion when students can speak out what they have learned and thought. Finally, they are expected to submit a report on the impression of this collaborative learning activity. (Table 1) Table 1. Sequence of the activities

Time Week 1

Activity Content Preparations including registration, adding applications, learning game rules and training basic operations.

Week 2 to Week 3 Make plans of farm managing and game scheduling

Week 4 to Week 6

Week 7

Enjoy the game according to the managing plan and game scheduling Discussion & Submit a report

Expected Capability information technology literacy, visual literacy analysis , team collaboration and organizational capacity, communication skills, information management capacity, knowledge development and management capabilities synthesis, critical thinking, self-control, problem-solving ability, decision-making ability, business intelligence evaluation, self-examination, innovation and creativity,

3.5 Roles Distribution The collaborative learning is a kind of constructivism activity. Teachers are conductors and helpers rather than tellers and controllers. They need to conduct the students to understand the game rules and choose partners to make a plan of farm managing together. When students come across troubles, they should not directly tell the answer but help them analyze problems. Students are no longer totally controlled as in traditional classrooms. They should make learning plans by themselves and discuss questions with partners. Before starting the game, they must understand game rules whereby the farm managing plan is reasonable. When playing the game, they need to execute the plan under strict self-control and modify it when necessary. After the game, they should be active self-examiners to review the process of the collaborative learning and summarize what they have learned and thought during the process.

Webgame Based Collaborative Learning Design: A Case Study

233

3.6 Type of Representation The types of representation include textual, graphical, video, audio, or oral representations of explicit instructions.

4 Conclusions This paper has provided an idea of innovation learning design for teachers to reuse webgames. From the analysis, it is clear that usability principles have been applied in the design of the Happy-Farm's interface. This will result in more effective selection of webgames. Only then can students reap the full benefits of the theories and tutorials without being distracted by poor usability design. The game can cultivate learners’ interests and motivation as it can provide an enrichment learning environment. The meaningful learning topic makes the game go beyond pure entertainment. In this collaborative learning design, learning is embedded in playing games and learners’ higher order thinking skills, such as abilities of making a plan, team spirit, self-control, self-examination and so on, have been developed.

References 1. Trilling, B., Hood, P.: Learning, Technology, and Education Reform in the Knowledge-edge or “We’re Wired, Webbed and Windowed, Now What? Educational Technology 39(5-6), 5–18 (1999) 2. Dillenbourg, P.: Collaborative Learning: Cognitive and Computational Approaches. Advances in Learning and Instruction Series. Elsevier Science, Inc., New York (1999) 3. Mitnik, R., Recabarren, M., Nussbaum, M., Soto, A.: Collaborative Robotic Instruction: A Graph Teaching Experience. Computers & Education 53(2), 330–342 (2009) 4. Lee, C.D., Smagorinsky, P. (eds.): Vygotskian perspectives on literacy research: Constructing meaning through collaborative inquiry. Cambridge University Press, Cambridge (2000) 5. Raskin, J.: The Humane Interface: New Directions for Designing Interactive Systems. Addison Wesley, Reading (2000) 6. Schneiderman, B.: Designing the User Interface, Strategies for Effective Human-Computer Interaction, 3rd edn. Addison-Wesley, Boston (1998) 7. Chiasson, S., Gutwin, C.: Design Principles for Children’s Technology (2005) 8. Dunham, T., Sindhvad, S.: Exploring developing and design of web-based learning environment for children (2005) 9. Hanna, L., Risden, K., Cserwinski, M., Alexander, K.J.: The role of usability research in designing children’s computer products. In: Druid, A. (ed.) The Design of Children’s Technology. Morgan Kaufman Publishers, San Francisco (1999) 10. Sullivan, T., Norris, C., Soloway, E., Peet, M.: When kids use the Web: a naturalistic comparison of children’s navigation behavior and subjective preferences on two. In: 6th Conference on Human Factor and the Web, Austin, TX (2000), http://www.sites 11. Gilutz, S., Nielsen, J.: Usability of Websites for Children: 70 Design Guidelines. Nielsen Norman Group, San Carlos (2000) 12. Chandler, D.: Visual perception 7 (1997)

234

J. Jian et al.

13. Helander, M.G., Landauer, T.K., Prabhu, P. (eds.): Handbook of Human-Computer Interaction, 2nd edn. North Holland, Amsterdam (1997) 14. Kruse, K.: E-learning and the neglect of user interface design 15. Bernard, M.: Optimal web design (2003) 16. Dillenbourg, P., Tchounikine, P.: Flexibility in Macro-Scripts for Computer-Supported Collaborative Learning. Journal of Computer Assisted Learning 23(1), 1–13 (2007) 17. Kollar, I., Fischer, F., Hesse, F.: Collaboration Scripts–A Conceptual Analysis. Educational Psychology Review 18(2), 159–185 (2006)

Design of a Medical Simulator Hard- and Software Architecture P. Peters, F. Delbressine, and L. Feijs Eindhoven University of Technology, Department of Industrial Design, Designed Intelligence group, The Netherlands [email protected]

Abstract. Using simulators for training is an accepted practice in medical education and in advanced medical training. Creating simulators that perform the functionality required and respond to interventions in a realistic way is key. The first iteration in the design cycle of creating a hard- and software platform that will support the development of these kind of simulators is the topic of this paper. The design approach, the hard- and software choices, the hard- and software architectures and the first results of creating a baby simulator prototype will be discussed. Keywords: Medical simulation, medical training, manikin.

1 Introduction In medical health care simulation training is one of the factors that help improving performance ([2, 5, 6]). Some findings show that using high-fidelity simulators offers additional training benefits ([5]), other findings show almost the opposite ([14]). The majority of the findings show that training by simulation is beneficial for basic education as well as for advanced training. This paper describes the first iteration of a design cycle performed to create a basic hard- and software architecture for human patient simulators to be used in baby delivery training. Different prototyping platforms have been evaluated ([13]) and a choice has been made for an implementation platform. The final results as used in the first prototype are shown in detail. The basic system as described in this paper is part of a complex system that allows to perform delivery simulation trainings (see figure 1). The patient simulators -or manikins- are only a small, albeit important part of this training system. The basic system, consisting of the hard and software of the simulators, is yet again only a small part of these simulators. 1.1 Context In the simulation center of the Máxima Medical Center in Veldhoven trainings for medical interventions will be performed. Projects are started to develop the simulation environments needed. One of these projects is focused on human patient simulators (HPS). In cooperation with the departments of Electrical Engineering and Industrial Design of the Eindhoven University of Technology a specific extra focus is on X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 235–246, 2010. © Springer-Verlag Berlin Heidelberg 2010

236

P. Peters, F. Delbressine, and L. Feijs

Fig. 1. Simulation training system

Fig. 2. Design process spiral model

simulators for deliveries. Research is done that enables to come to a realistic simulation of a delivery. Specific attention is given to simulators for mother and child as well as the methods to monitor the performance of medical staff. Although this paper focuses on development of a prototype for a neonate baby manikin, the results will also be usable in the mother manikin since the basic system elements are the same. 1.2 Design Process The design process followed is a spiral model, combining elements of design and prototyping-in-stages, combining advantages of top-down and bottom-up concepts as proposed in [1]. This process is combined with ideas from the software/hardware design process as described in [12]. The result is an iterative process including user and technology requirements research,concept design, prototype implementation and validation as shown in figure 2. Analogous to the process of the MASTER methodology ([7]) used in military simulator-based training, where simulator and training are designed in 3 related steps: training needs analysis, training programme design, and

Design of a Medical Simulator Hard- and Software Architecture

237

training media specifications, the requirements for the simulator will be a direct result of the training needs analysis conducted by consulting expert medical staff. The inquiry will cover medical skills requirements and training needs for each assisting staff member as well as skill requirements and training needs for the team as a whole. Via the medical skill requirements and training needs and the resulting requirements for system functionalities, the parameters required to measure (e.g. force, pressure, position) and the required feedback (e.g. skin color, heart rate, blood pressure) can be determined. These functionalities and responses will be realized in -possibly partialprototypes that can be tested. This results in an analysis and prototyping process as depicted in figure 3. The feedback of the medical staff that uses the prototype will be taken into account when building the next prototype. Using iterative design enables evaluation and to increase performance in the next iteration. New ideas might pop up that can be implemented. This easily leads to an amount of work that is too much for one research group to handle. Therefore an open architecture and open design (similar to open standards/open source software) will be used such that contribution of other research and development groups is possible.

Fig. 3. Needs analysis to prototype

1.3 Design Process Implications Since not all functionality required for all interventions can and have to be implemented at once, choices are made. This allows choices for sensors and actuators used for a specific functionality to be delayed until the time comes this functionality actually is implemented. The basic system that handles these sensors and actuators needs to allow for adding newly developed sensors and actuators and for implementation of new functionalities in the overall system; it needs to be flexible and allow for extension. The basic system needs to be developed in such a way that it survives several iterations. 1.4 Design Considerations The goals chosen for this project are two-fold. First, intermediate results of the project need to be realized quickly and should be evaluated. Second, the sum of all

238

P. Peters, F. Delbressine, and L. Feijs

intermediate results should lead to a complete hard/software architecture that can be used in medical simulators. Although described differently, the levels of design as described in [11] are visible here as well. The hardware and software design for the manikin prototype will be the result of the requirements that come out of the training needs analysis and the exercise design.

2 Requirements Besides selecting sensors to measure the required physical parameters and actuators to give appropriate feedback, the basic system handling sensor input and actuator output is designed. The requirement that new functionality and more sensors and actuators can be added demands an implementations highly independent of type and amount of sensors and actuators. One way to do that is to communicate sensor values read to the main system and to receive values from the main system to control the actuators. This can be implemented using a simple micro-controller (see section 3). 2.1 Body Parameters Only a small subset of all bodily parameters is of importance for actual simulation.Examples are shown in table 1 (source [15] and extended). Although to the human body these parameters are normally internal or output parameters, in the simulator these parametersmight be input as well as output. For instance, chest compression rate values needed in CPR will be derived from the available heart-rate values. Table 1. Medical and physiological parameters

Parameter Blood flow Arterial blood pressure

Range 1-300 ml/s

Frequency dc-20 Hz

Sensor Flowmeter

10-400 mmHg

dc-50 Hz

Respiratory flow

0-600 l/min

dc-40 Hz

Respiratory rate

2-50 breaths/min

0.1-10 Hz

Heartbeat

30-300 bpm

0.5-5 Hz

Muscle reflex

δt = 0.255 s

2-100 Hz

Strain gage manometer Pneumotachograph Strain gage, Impedance, Nasal thermistor Skin electrodes, Stethoscope Timer

2.2 Physiological Models There are 2 distinct approaches to let a manikin autonomously simulate responses, simulations scripts and models ([11]). Simulation scripts interpreted by the scenario

Design of a Medical Simulator Hard- and Software Architecture

239

computer (see figure 1) result in commands sent to the manikins and thus in actions performed, based on sensor information (from the manikin or environment).Complex models of physiological processes (e.g. feto-maternal blood circulation ([8])) can be implemented, controlling the manikin behavior autonomously. Whether these models will be in the main computer (see figure 1), the manikin, or are distributed is unknown at this point in time. 2.3 Physical Restraints When creating a neonate baby manikin the amount of space for all electronic and mechanic elements is very limited.

(a) Side view

(b) Frontal view

Fig. 4. Manikin measurements

Taking measurements from figures 4(a) and 4(b), space would be approximately 615 cm3 for the body and 244 cm3 for the head. The size constraint makes it necessary to think about miniaturization. Placement of components needs attention too. Some sensors and/or actuators need to be at a specific locations; other locations are not suited to place certain components, e.g. placement of components at a joint, disabling the joint to bend, is undesirable. Weight of the components also plays a role. The total weight of the manikin has to be realistic and, dependent on the type of baby manikin required, is in the range of 500 grams to 4000 grams. These restraints affect factors like electric power (battery size) and actuator force (motor size).

3 Implementation 3.1 Hardware Implementation As mentioned in section 2, the basic system setup allows for many types of sensors and actuators, although the prototype implementation described will only contain a few. Since the complete simulation training system is not available yet, a test environment is used to test the functionality of the basic system. A diagram of the basic system and test environment is shown in figure 5.

240

P. Peters, F. Delbressine, and L. Feijs

Fig. 5. Basic system and test environment

The functionality of the first prototype is limited, to limit complexity and to enable immediate implementation, without need for a complete requirements and functionality specification. Medical staff indicated that training for forceps delivery and training for specific movements during normal delivery is highly likely. With this information a partial needs-analysis is created (see figure 6) and part of this functionality is implemented in the first prototype.

Fig. 6. First needs analysis diagram

The functionality chosen to be implemented is: • • •

Measuring the baby’s chest compression depth and rate. Measuring position/rotation of the baby. Measuring pressure on the baby’s head.

Chest compression information can be used to assess correct execution of CPR. Rotation of the unborn child is information valuable to assess correctly performing several manoeuvres used in delivery like the Løvset manoeuvre or Bickenbach manoeuvre. Measuring the pressure where the forceps grab the baby’s head gives a direct indication of the force applied.

Design of a Medical Simulator Hard- and Software Architecture

241

Sensors Sensors in the manikin prototypes will be read using A/D conversionl, or by directly reading digital input ports in case of a digital signal. Measuring rotation of the baby manikin is done by a dual axis micro- electromechanical system (MEMS) acceleration sensor. The sensor chosen is a low power dual axis accelerometer (Analog Devices ADXL311) that converts acceleration into an electrical signal in the range of 1.15 to 1.85 Volts1. This sensor can also measure static acceleration, such as gravity, allowing it to be used as a tilt sensor. An amplifier / buffer circuit is added that extends the output range from 1.0 to 2.0 Volts1 (see figure 7).

Fig. 7. Acceleration sensor schematics

The sensor that enables detection of chest compression rate and depth is a piëzoresistive pressure sensor. This pressure sensor is connected to a flexible air-tight chamber located at, or close to the heart position. The compression of the flexible chamber is proportional to the chest compression applied. The chest compression rate can be measured counting the number of compressions during a specific time interval. The sensor chosen (Honeywell 40PC001B2A) is able to directly convert a pressure range of -50 mmHg to +50 mmHg into an electrical signal in the range of 0.5 to 4.5 Volts2 at a sensitivity of 40 mV/mmHg typical. Actuators The actuators in the manikin prototype will convert electrical signals to specific physical values. These signals will be created by the micro controller, which offers only digital I/O (switching between approximately 0 and 5 Volts). The micro controller offers the possibility to use this to create pulse width modulation (PWM) which enables a kind of continuous output using digital I/O. Additional circuitry can be added to do actual digital to analogue (D/A) output enabling real analogue output signals. Actual test implementations of stepper motor functionality have been made and proven to be possible, as well as generation of heartbeat sounds using a small

242

P. Peters, F. Delbressine, and L. Feijs

speaker. The created prototype does not contain these functionalities, and although not used, the choices for output signals driving the actuators are made based on these experiences. 3.2 Micro Controller The micro controller takes care of converting sensor signals to data and transferring this data in messages to the main system. In the opposite direction, messages sent by the main system to the micro controller contain data used to control actuators. For the first prototype a choice are is made for analog and digital input reading sensors and for digital output driving actuators The micro controller chosen (Microchip 18F4550) has possibilities for doing 10 bit A/D conversion on 13 channels (sequentially), performing digital input, performing digital output as well as PWM, furthermore it has rudimentary provisions for asynchronous serial communication (RS232) as well as for synchronous serial communication (I2C, SPI) and USB. A ZigBee®; communication module connected to the micro controller provides it with a wireless serial communication channel. Data is transported over this channel using a standard character stuffing protocol explained in section 3.3. Power Consumption In the prototype created, attention to minimize power consumption is done by choosing low power sensors and choosing a micro controller and a communication module that can be put into sleep mode. The total power consumption is approximately 290 mW. The battery used to power the prototype is a 3.6 Volts lithium-ion battery pack of 4.1 Wh. Fully charged this will power the prototype for about 8.5 hours taking into consideration that a Li-Ion battery will start dropping voltage at 60% of it’s capacity. 3 possibilities exist to get the battery pack charged: 1- replacing the battery pack with a charged one (that has been externally charged), 2- charging the battery pack inside the manikin via a built in connector, 3- charging the battery pack inside the manikin wirelessly via induction ([3],[10]), the last method has preference. 3.3 Software Implementation General The software in the baby manikin takes care of reading sensors, driving actuators, doing conversions, timing relations and communication to the communication server (see figure 1). Although the actual implementation of the software is still simple compared to what it will be, the software architecture is according to the one depicted in figure 8 and the same architecture will be used for future prototypes. The manikin software is divided into layers that perform specific functions like physical interfacing, abstracting, providing logical services and the actual manikin functionality. Every layer communicates with neighboring layers above and below. The connection between the manikin and rest of the system (represented as ”PC” block in figure 8) is done via the physical layer. From a logical point of view, layers with same names communicate with each other as symbolized with the dotted lines.

Design of a Medical Simulator Hard- and Software Architecture

243

Fig. 8. Software architecture

Programming The language used to program the micro controller is Microchip MPLAB®; C18. The main program consists of a loop reading 8 analog input sensors, collecting available incoming serial data, performing filtering on the available input data, checking if output has to be done and performing output data transmission. Serial data is received based on interrupt, and time critical actions are performed based on a 1mS timer interrupt. Incoming data messages are used to configure the software and to control actuators. The only configurable feature in this prototype is the selection to send messages containing the sensor data on request or once every 20mS. Sensors All connected sensors are read sequentially, data values are stored and the mean value of the last 10 measurements is calculated and used as data to send in the sensor message. This calculation creates a low pass filter that reduces the amount of noise that might be present on the signal but it also prevents quick changes to be detected and introduces delays. In the first prototype the delays were acceptable, but the filtering and delays are certainly undesirable in case of high speed signals (> 5 Hz) or time critical related signals. Because these filtering calculations take time, they also put a limit on the amount of sensors that can be processed. Actuators Actuators are driven as soon as the type of the incoming message indicates it is an actuator message. The message will also contain an indication of which actuator to drive and a value that determines what to do with the actuator. In case of e.g. the heartbeat generator it could contain the number of heartbeats per minute, in case of a stepper motor it could contain direction, step speed and number of steps. Timing The micro controller platform used in this prototype allows for reading 8 sensors (and driving 8 actuators) with a repetition rate of 20 mS, which is more than enough for the sensors and actuators implemented. The muscle reflex signal in table 1 is the signal that requires the fastest reaction time of the micro controller. For signals of this kind,

244

P. Peters, F. Delbressine, and L. Feijs

a sensor with local processing capability, a sensor with direct coupling to the related actuator, or possibly a more powerfull micro controller might be needed. Communication In the final system, the baby manikin communicates with the communication server (see figure 1).For now a communication stub is used. This stub -a program running on a PC- simulates the rest of the final system as far as it is of importance to test the prototype. For different purposes (debugging, testing, demonstration) several of these stubs have been created in varying programming languages (Java, Max/Msp, Processing). Data Protocol The protocol used for data transportation consists of 2 layers. In the terminology of the ISO OSI protocol ([16]) the Physical link layer and the Media access control layer (a sublayer of the Data link layer) are implemented. The Physical link layer uses a standard RS232 communication protocol.

Fig. 9. Data packet

The layer on top of that encodes the data into byte messages layed out like in figure 9. Encoding of data is done using a character stuffing protocol. For integrity checking, correction and addressing a cyclic redundancy check (CRC), a message sequence number and an ID is added to the message.

4 Evaluation The baby manikin prototype discussed is the first one of a series. This particular prototype was used to gain knowledge about possible problems in hard and software development as well as problems in actually building the hardware and software into a real life-like sized human baby model. The finalized first prototype was successfully used in a baby delivery demo, performed at an event organized to celebrate the successful end of the Stimulus project that funded this work. At the start of this project, a lot of requirements for the baby manikin were unknown, and even at the current stage, there are many unknown factors. The process of gathering scenarios for training and deriving requirements for the manikins is still going on. Yet this exercise already has brought new knowledge in several areas: possibilities and limitations of the chosen platform, available sensors usable for the baby manikin, things to do and not to do when actually building the hardware into a baby-model. 4.1 Future Work Reflecting on the experiences, designing, building and testing the first prototype, some considerations for future work come to mind:

Design of a Medical Simulator Hard- and Software Architecture

245

• Implementation of more functionalities: multiple analogue and digital sensors, motor control and audio output, might require more programming memory and random access memory than available now; especially if the amount of sensors/actuators is increased a lot. • The timing used for sensor reading, the speed of the serial transmission and the filtering done on the sensor values prevent high speed sensor changes. Several solutions to this problem are possible, ranging from changing the software implementation to changing the hardware platform to a more powerful and faster one. • The power consumption of the manikin will increase significantly when adding actuators. Motors or similar components tend to consume a relatively high amount of power (compared to the sensors and micro controller). • The data communication protocols implemented are still low-level. In the final system there will probably be at least one more protocol layer (XML is a likely candidate [4, 9]). • Assumptions are made about the interface between the baby manikin and the rest of the system. Although educated guesses can be made on some possibilities e.g. what physical parameters to measure, types of sensors and actuators to use, others like types and frequency of events to be measured, types and frequency of feedback to be generated- are still unknown, and yet others -like the scenarios- are likely to be extended or changed. • Tests with a larger number of users should be performed to evaluate the functionality and realism of the manikin prototype.

Acknowledgements The author would like to thank prof.dr.ir. Loe Feijs and dr.ir. Frank Delbressine, both member of the Designed Intelligence group of the department of Industrial Design at the Eindhoven University of Technology. They have taken the time and put in the effort of reading and commenting on all versions of this paper while it was being written.

References 1. Boehm, B.W.: A spiral model of software development and enhancement. SIGSOFT Softw. Eng. Notes 11(4), 14–24 (1986) 2. Burchard, E.R., Lockrow, E.G., Zahn, C.M., Dunlow, S.G., Satin, A.J.: Simulation training improves resident performance in operative hysteroscopic resection techniques. American Journal of Obstetrics and Gynecology 197, 542.e1–542.e4 (2007) 3. Chen, W., Sonntag, C., Boesten, F., Bambang Oetomo, S., Feijs, L.M.G.: A power supply design of body sensor networks for health monitoring of neonates. In: Proceedings of the 4th International Conference on Intelligent Sensors, Sensor Networks and Information Processing, ISSNIP 2008, pp. 255–260. IEEE, New York (2008) 4. Chen, W., Hu, J., Bouwstra, S., Oetomo, S.B., Feijs, L.: Sensor Integration for Perinatology Research. Journal of Sensor networks (to appear, 2010) 5. Crofts, J.F., Bartlett, C., Ellis, D., Hunt, L.P., Fox, R., Draycott, T.J.: Training for shoulder dystocia: A trial of simulation using low-fidelity and high-fidelity mannequins. Obstetrics and Gynecology 108, 1477–1485 (2006)

246

P. Peters, F. Delbressine, and L. Feijs

6. Deering, S., Brown, J., Hodor, J., Satin, A.J.: Simulation training and resident performance of singleton vaginal breech delivery. Obstetrics & Gynecology 107, 86–89 (2006) 7. Farmer, E., van Rooij, J., Riemersma, J., Jorna, P., Moraal, J.: Handbook of SimulatorBased Training. Ashgate Publishing Ltd., Aldershot (1999) ISBN 0754611876 8. van der Hout, B.: Development of a simulation model of the fetomaternal circulation. In: Oral presentation at 14th Annual Meeting of the Society in Europe for Simulation Applied to Medicine, June 19-21 (2008) 9. Hu, J., Feijs, L.: A Distributed Multi-agent Architecture in Simulation Based Medical Training. In: Chen, Q. (ed.) Transactions on Edutainment III. LNCS, vol. 5940, pp. 105–115. Springer, Heidelberg (2009) 10. Ma, G., Yan, G., He, X.: Power transmission for gastrointestinal microsystems using inductive coupling. Physiological Measurement 28, 9–18 (2007) 11. van Meurs, W., Good, M., Lampotang, S.: Functional Anatomy of Full-Scale Patient Simulators. Journal of Clinical Monitoring and Computing 13, 317–324 (1997) 12. Overmyer, S.P.: DoD-Std-2167A and methodologies. SIGSOFT Software Engineering Notes 15(5), 50–59 (1990) 13. Peters, P., Feijs, L., Oei, G.: Plug and Play Architectures for Rapid Development of Medical Simulation manikins. In: Proceedings of WMSCI 2008 - The 12th World MultiConference on Systemics, Cybernetics and Informatics: WMSCI 2008, Orlando, Florida, USA, June 29-July 2, vol. 2, pp. 214–219 (2008) 14. Scerbo, M.W., Dawson, S.: High fidelity, high performance? Simulation in Health care 2, 224–230 (2007) 15. Webster, J.G.: Medical Instrumentation - Application and design, 3rd edn. John Wiley & Sons Inc., Hoboken (1998) ISBN 0471153680 16. Zimmerman, H.: OSI Reference Model - The ISO Model of Architecture for Open Systems Interconnection. IEEE Transactions on communicaions COM-28(4), 214–219 (1980)

Design and Implementation of Semantic Matching Based Automatic Scoring System for C Programming Language Jinrong Li, Wei Pan, Ren Zhang, Feiquan Chen, Shenglong Nie, and Xiaoming He∗ School of Software, Northeast Normal University Changchun 130117, China [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]

Abstract. With the development of education and computer, it has become an urgent demand for scoring programming questions automatically. This paper puts forward a semantic matching based automatic scoring method for C programming questions. The said method standardizes student programs and template programs, and then calculates their semantic similarity, so as to score student programs. Main idea: Firstly, convert student programs and template programs into an intermediate representation of programs (i.e. system dependence graph); secondly, carry out semantic equivalence conversion for the generated system dependence graph according to a series of standardization rules, so as to eliminate the diversification of program expression; thirdly, calculate the matching degree of the standardized system dependence graph, and score student programs according to the matching result and scoring rule. Keywords: Automatic scoring, semantic matching, program standardization.

1 Introduction With more and more extensive application of computer, examinations on “programming languages” are becoming more and more common. Therefore, network teaching and automatic examination of “programming languages” have become practical research objects. At present, computer marking has been widely used in various examinations. Marking test papers automatically by using computer not only can greatly improve work efficiency and prevent incorrect marking, but also can eliminate the influence of artificial factor and realize the objectiveness and fairness of examination. Among various kinds of computer marking studies, automatic examination of “programming languages” is the most urgent one. [1] If programming questions can be automatically scored by using computer, it not only can free teachers from boring and tedious marking work, but also can make scores of the students more fair and reasonable. This paper proposes a semantic matching based automatic scoring model for programming questions. The said model scores student programs by simulating the ∗

This work was supported by National Collegiate Innovative Experiment Program.

X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 247–257, 2010. © Springer-Verlag Berlin Heidelberg 2010

248

J. Li et al.

thinking process of artificial marking, and through comparing student programs with template programs. Main processes are as follows: 1. Make lexical and grammatical analysis for the source program, and meanwhile, convert the program into system dependence graph; 2. Take a series of standardized processing methods on the basis of system dependence graph to eliminate the diversification of program expression; 3. Match student programs with template programs from four levels, namely program size, structure, depth and knowledge application, and score the program according to the similarity of the matching result and the scoring criteria. This paper introduces in turn the key and difficult point of evaluating student programs by using computer, as well as the proposing and realizing process of automatic scoring model.

2 Model Research 2.1 Features of Student Programs This paper focuses on the programs made by students when taking programming language examinations. There are many advantages of the evaluation of these programs over the understanding of other types of programs: 1. It indicates explicitly in the question of each programming question that students must make a program realizing a fixed function. Therefore, objective of the program is very clear, and its function is completely described; 2. Only a small quantity of codes are required, generally not more than 50 lines, and most of which are between 10 lines and 30 lines; 3. Programming questions generally check the degree of understanding and the ability of applying certain knowledge point of a programming language, so these programs are of simple function and involve less knowledge points; 4. Keys to many programming questions are of fixed structures, so they are more suitable for the computer to understand; 5. Limited by their ability, most students won’t use more advanced knowledge or skill during programming, and they basically use programming methods taught by the teacher or in the textbook. Meanwhile, programming questions mainly check the understanding and application of basic knowledge in the textbook, and seldom involve comprehensive or very difficult questions. [2] Therefore, this paper suggests that it is available to evaluate student programs automatically by using computer, and attempts to put forward a scheme for programming questions and computer scoring. 2.2 Difficulties to Be Resolved There may be two kinds of mistakes in student programs, namely grammar mistake, and mistake of applying semantics and program knowledge & concept. [3] When

Design and Implementation of Semantic Matching Based Automatic Scoring System

249

marking programming questions artificially, it mainly takes the quantity and type of each kind of mistake as the scoring criteria. Sufficient theoretical basis for grammatical analysis has been provided by compilation technology, so it is very easy to implement grammar mistake detection. However, there is no good solution for semantic mistake detection yet. For the above reasons, in order to score programming questions automatically, the following problems should also be solved besides the problem of general grammar mistakes: 1. How to deal with student programs with diversified expression; 2. How to deal with incorrect programs or incomplete programs of students; 3. Criteria used for scoring. 2.3 Proposing the Model In order to solve the above problems, the thinking process of artificial scoring should be studied first. Before artificial scoring, there are one or more standard keys to each programming question. These standard keys use different problem-solving methods. Served as different ways to solve one programming question, such standard keys are correct and optimized. During scoring, teachers may firstly find out the most suitable key by comparing the student program with standard keys, and then analyze the programming details, with the purpose of scoring the student program according to the matching degree. If there is grammar mistake in student programs, some corresponding scores will be deducted additionally. Therefore, difficulties of automatic scoring can be solved through simulating the artificial scoring method and combining features of handling problems by computer: 1. Keep semantics of the program unchanged, and standardize the program using a series of rules, so as to eliminate diversification during implementing programs; [4,5] 2. Incorrect or incomplete student programs can’t completely contain knowledge points and concepts involved in the question, but the similarity of the semantic function between each statement of student programs and standard keys can still be known through program matching, so as to solve the scoring problem of incorrect or incomplete programs; 3. This information can be converted into detailed scores by combining grammar detection information and referring to artificial scoring rules on the basis of similarity matching of student programs and template programs. As for the problems to be solved for scoring programming questions automatically, this paper finds out the best code form for similarity detection, i.e. system dependence graph. Combining the features of student programs, this paper proposes a semantic matching based automatic scoring method for C programming questions. The model of this method mainly consists of the following three stages: 1. From making lexical and grammatical analysis on the source code to the conversion of system dependence graph; 2. System dependence graph based program standardization;

250

J. Li et al.

3. Scoring by matching standardized student programs with template programs. [6,7,8]

3 Model Implementation 3.1 Lexical and Grammatical Analysis Lexical and grammatical analysis should be made on student programs and template programs before the conversion of system dependence graph. Lexical and grammatical analysis mainly applies the relevant theories of compilation technology, but they are different in task partitioning. In the automatic scoring model, main task of lexical analysis is to form TOKEN string and error handling, but not involve symbol table searching and filling; while grammatical analysis is mainly to analyze variable declaration statement, identify limits of each function of the program, and search and fill in the symbol table. There is a detailed description of this method in the compilation technology, so it will not be described here. [9,10] 3.2 Conversion of System Dependence Graph System dependence graph [11] is a common representation during analyzing programs; it can clearly show various control dependence relations and data dependence relations of the programs. Comparing with flow diagram method of programs, it is easier to convert the program into system dependence graph, and the representing result is concise and clear, so system dependence graph is taken as the program representation in this paper. A system dependence graph (Gp) of program (P) is an identified and directed multivariate graph, each node of which represents a program concept, such as declaration statement, assignment statement, and control predicate. [12,13,14] Besides, there are two special nodes in the graph, namely entry node corresponding to each function or process, and head node corresponding to the entire program. Each edge of Gp represents various relations between nodes, and these relations are generally distinguished by the identification of the edge. Relations (types of the edge) between nodes of the system dependence graph are basically divided into three categories: Control relation, data relation, and declaration relation. According to the definition of system dependence graph and various dependence relations, and by combining the statement types of C programming language, the types of the nodes in the system dependence graph of C programs are mainly divided into declaration node, assignment node, function call node, increment/decrement node, return node, continue node, break node, if/if-else node, switch node, do-while node, while node, for node, and head node and entry node, which represent program startup and function entry respectively. [2] Besides type identification information, these nodes should record the beginning and end position of statements corresponding to the node in Token string, the respective expression content, control flow dependence node and data flow dependence node. The following figure is the system dependence graph of a simple program.

Design and Implementation of Semantic Matching Based Automatic Scoring System

251

int main() { int i, s = 0; for (i = 1; i P riority[C] > P riority[B]. Finally, the communication priority tree is built as shown in Fig. 2. The dotted lines with arrows in Fig. 2 represents the priorities of the childnodes descend in turn along the direction of the arrows.

304

H. Huang et al.

After the destination node group is selected on basis of communication delay, the destination node of the migrating federate f can be determined in the destination group according to the following formula similar to (4): DestN ode(f, gt ) |g |

t = {pj |pj = M ink=1 {LoadV ar(f, pk , gt )}} .

In the algorithm for selecting the migration destination node among the groups above, the computation of communication delays of the logical node groups can be implemented “offline” before the emergence of federate migration event, and the system should maintain and update the communication delay information of every logical node groups. When the federate migration event arrive at the inter-group load management module, the communication delay data of logical node groups gs and gt can be extracted directly so as to improve the speed of response of inter-group migration decision.

4

Performance Evaluation

We have realized a federate migration management system, and carried out comparison test with random load balance algorithm. In the test, the running environment of simulation grid is composed of 5 host computers. The information service runs on computing node p2 . The management module for every node monitors the current load of it, and submits the information to information service periodically. HelloWorld federate is adopted. When the simulation federation is built, more HelloWorld federates run on computing node p1 by design to maintain higher load of p1 for convenience. In the process of federation running, suppose federate migration event will be triggered when the load of computing node p1 exceeds ( = 50), that is to migrate HelloWorld federates on p1 to other computing nodes. Each time one HelloWorld federate will be migrated, until the load of p1 is less than or equal to 40. The test is consists of two sets (12 independent experiments), the first set (the former 6 independent experiments) adopts the algorithm proposed in this paper to compare the load changes of the nodes before/after federate migration, and the second set (the latter 6 independent experiments) compares the algorithm to random selection algorithm, and computes the load unbalance values of all 5 computing nodes on basis of formula u,v=1 |Lc [u] − Lc [v]|/2. The comparison curves of load unbalance values before/after federate migration of the first set of experiments are shown in Fig. 3. Fig. 3 shows that the load balance conditions of the system are improved obviously after federate migration. The second set of experiments adopts both the algorithm proposed in this paper and the random selection algorithm for federate migration. And the curves of load unbalance values are shown in Fig. 4. Fig. 3 and Fig. 4 show that the load unbalance value changes more obviously by adopting random selection algorithm, while it changes relatively stably and

Node Localization for Distributed Simulation Based on Logical Node Group load unbalance value before federate migration load unbalance value after federate migration

250

load unbalance value

305

200

150

100

50

0 1

2

3

4

5

6

experiment number

Fig. 3. Changes in load unbalance value before/after federate migration 120 proposed algorithm random algorithm

load unbalance value

100

80

60

40

20

0 1

2

3

4

5

6

experiment number

Fig. 4. Changes in load unbalance value adopting different localization algorithms

maintains at a smaller level by adopting the algorithm proposed in this paper. Compared to the random selection algorithm for federate migration, the proposed algorithm shows advantages in improving system load balance.

5

Conclusions and Future Work

In a distributed simulation running, the initial “task-computing node” mapping should change with grid system resources. A delay-centered node localization algorithm for federate migration in simulation grid on basis of logical node group is proposed. When the destination node of the federate is located, the load of computing node, the network communication bandwidth, and differences of communication delay are all considered. The average communication delay between federates is reduced while the load system realizes load balance. A federate migration system is realized to prove the efficiency of the proposed

306

H. Huang et al.

algorithm in achieving load balance. In later research, more comparison tests will be implemented for the performance indicator of reducing average message communication delay. Acknowledgments. This work is supported by the National 863 Program of China (No. 2009AA01Z333), National Fundamental Research 973 Program of China (No. 2009CB320805), China Postdoctoral Science Foundation (No. 20090450280), Beijing Natural Science Foundation (No. 9103023), Fundamental Research Funds for the Central Universities (No. BUPT2009RC1016), and National Natural Science Foundation of China (No. 60702049).

References 1. Zajac, K., Tirado-Ramos, A., Zhao, Z., Sloot, P., Bubak, M.: Grid Services for HLA-based Distributed Simulation Frameworks. In: Fern´ andez Rivera, F., Bubak, M., G´ omez Tato, A., Doallo, R. (eds.) Across Grids 2003. LNCS, vol. 2970, pp. 147–154. Springer, Heidelberg (2004) 2. Xie, Y., Teo, Y.M., Cai, W., Turner, S.J.: Extending HLAs Interoperability and Reusability to the Grid. In: 19th ACM/IEEE/SCS Workshop on Principles of Advanced and Distributed Simulation, USA (2005) 3. Hyett, M., Wuerfel, R.: Connectionless Mode and User Defined DDM in RTI-NG V6. In: The Proceedings of the 2003 Spring Simulation Interoperability Workshop (2003) 4. Yuan, Z.J.: Load Management System for Distributed Simulation. Division of Computer Science, School of Computer Engineering, Nanyang Technological University, Singapore (2003) 5. Simulation Interoperability Standards Committee (SISC) of the IEEE Computer Society. IEEE Standard for Modeling and Simulation (M&S) High Level Architecture (HLA) - IEEE std 1516-2000, 1516.1-2000, 1516.2-2000. The Institute of Electrical and Electronics Engineers Inc., New York (2000) 6. Pullen, J.M., Brunton, R., Drake, D., et al.: Using Web Services to Integrate Heterogeneous Simulations in a Grid Environment. In: Bubak, M., van Albada, G.D., Sloot, P.M.A., Dongarra, J. (eds.) ICCS 2004. LNCS, vol. 3038, pp. 835– 847. Springer, Heidelberg (2004) 7. Brunton, R.P.Z., Morse, K.L., Drake, D.L., Moller, B., Karlsson, M.: Design Principles for a Grid-Based HLA Federation. In: Proceedings of the IEEE 2004 European Simulation Interoperability Workshop, Paper 04E-SIW-056 (2004) 8. Junwei, C., Spooner, D., Jarvis, S., et al.: Agent-Based Grid Load Balancing Using Performance-Driven Task Scheduling. In: Proceedings of the 17th International Parallel and Distributed Processing Symposium (2003) 9. Cao, J., Spooner, D.P., Jarvis, S.A., Nudd, G.R.: Grid Load Balancing Using Intelligent Agents. Future Generation Computer Systems 21(1), 135–149 (2005) 10. Leinberger, W., Karypis, G., Kumar, V., et al.: Load Balancing Arcoss NearHomogeneous Multi-Resource Servers. In: Proceedings of Heterogeneous Computing Workshop, pp. 60–71 (2000) 11. Lan, Z.: Dynamic Load Balancing for Parallel and Distributed Systems. Northwestern University (2002) 12. Arora, M.: DONAN: An Efficient Algorithm for De-centralized Scheduling and Resource Utilization in Grid Environments. The University of Texas at Arlington (2002)

Using Graph Edit Distance to Diagnose Student's Science Process Skill in Physics Ming-Xiang Fan1, Maiga Chang2, Rita Kuo3, and Jia-Sheng Heh1 1

Dept. of Information and Computer Engineering, Chung-Yuan Christian Univ., Taiwan 2 School of Computing and Information Systems, Athabasca University, Canada 3 Dept. Of Digital Design, Mingdao Univ., Taiwan [email protected], [email protected], [email protected], [email protected]

Abstract. Science process skill is important to students when they are learning sciences. We have built a story-based virtual experiment environment in Physics to train student’s science process skills. However, the teacher currently doesn’t know how the students’ science process skills are while they are using the virtual experiment environment. For this reason, this research tries to apply the graph edit distances to design the graph-based diagnosis methodology and uses the methodology to compare the student’s science process skill graph and the expected graph in order to give both the teacher and the student feedback regarding the student’s science process skills. This research also proposes the experiment design and plans to do the experiment in the end of this May (May, 2010) to prove the effectiveness of the proposed graph-based diagnosis methodology in analyzing students’ science process skills in Physics. Keywords: Science Process Skill, Virtual Experiment Environment, Storybased Virtual Experiment, Physics, Problem Solving, Graph edit distance, Diagnosis.

1 Introduction There are many researches use the virtual experiment environment to teach Science topics [4][5]. We have built a story-based virtual experiment environment in Physics to train students’ science process skills. However, the most important concern the teacher may have is how students’ science process skills change after they used the system. For this reason, this research wants to diagnose a student’s science process skills after s/he completed the virtual experiment and to give the teacher feedback with descriptive explanations of the student’s skills and quantitative score as the measurement of the skills. The virtual experiment environment we have built combines the science process skills and seven stages of problem solving [17]. Each problem solving stage has been mapped to different stages of doing virtual experiment. The student’s manipulation behavior at each stage then can be considered as his/her performance in specific science process skill. This research transforms the student’s manipulation behavior at X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 307–316, 2010. © Springer-Verlag Berlin Heidelberg 2010

308

M.-X. Fan et al.

each stage to a graph, and then uses the revised graph edit distance methodology to generate diagnosis result to the student and the teacher. Section 2 describes relevant research which we use to develop the virtual experiment environment and to design the graph-based diagnosis methodology, e.g. science process skills, seven stage of problem solving, and graph edit distance. Section 3 introduces the virtual experiment environment we have built. Section 4 explains the methodology of using graph edit distance to diagnose the student’s science process skill. In Section 5, we talks about the experiment design we are going to do in this Summer. At the end, Section 6 discusses possible future works.

2 Research Background This research designs a graph-based diagnosis methodology to diagnose students’ science process skills when they do Physics experiment in the virtual experiment environment. The virtual experiment environment [7][16] allows students doing experiments by using virtual equipments [4]. Experiment is a process of solving problem. Researchers, e.g. Polya (1957) and Mitchell and Kowalik (1989), have proposed different problem solving strategies [8][11]. Polya’s problem solving strategy involves four stages: understanding the problem, devising a plan, carrying out the plan, and looking back. Mitchell and Kowalik’s problem solving strategy has six stages: mess finding, data finding, problem finding, idea finding, solution finding, and acceptance finding. Seven Stages of Problem Solving

Science Process Skill

Problem Determination

Formulating hypotheses

Problem Transformation

Classify Variable identifying

Experiment Design & Planning

Variable identifying

Experiment Execution

Experimenting

Data Recording & Observation

Acquiring Data

Data & information explanation

Constructing Graphs Analyzing Investigations

Experiment Result Evaluation

Analyzing Investigations

Fig. 1. Revised relations between science process skills and the seven stages of problem solving based on Kuo et al. (2000)

The seven stages of problem solving can lead student to solve science problem while doing experiment [1]. Kuo and her colleagues (2000) have analyzed the relations between the seven stages of problem solving and the science process skills as Fig. 1 shows [5]. Students who follow the stages one by one to do Physics experiments can practice different science process skills at different stage. If they think they

Using Graph Edit Distance to Diagnose Student's Science Process Skill in Physics

309

have mistakes at one stage, they could go back to the previous stages. Each stage associates with different science process skills listed at the right hand side of Fig. 1. The story-based virtual experiment environment we have created can automatically generate different virtual experiment from a knowledge structure. Many researchers discuss how to construct the knowledge structure according to human’s mental model. For examples, Concept Map [10] is widely used by psychologists and educators; Knowledge Map [6] is also widely used in the e-learning field to store the learning contents. This research uses the Context-Aware Knowledge Structure which is proposed by Wu et al. (2008) to store the knowledge for generating virtual experiment. The context-aware knowledge structure [15] has three layers: 1. Domain layer: denotes the subjects and the topics of the learning environment. 2. Characteristic layer: stores the characteristics of the subjects and the topics. 3. Object layer: represents the real learning objects. As long as this research wants to design a graph-based diagnosis methodology, we need to have some definitions and symbols to represent graphs first. Graph is a useful data structure which can be used to represent various objects and concepts. Using graph can transform the behavior diagnosis problem to a graph matching problem. Edit distance is a method which can measure the difference between two data structures suchlike strings [14], trees [13] and graph [12]. Three operations in graph edit distance are insertion, deletion, and relabeling for both nodes and edges. The error-correcting graph matching [2] between two graph G1 = (V1 , E1 ) and G2 = (V2 , E2 ) has six graph edit operations for changing G1 to

G2 : substitute a node in G1 , delete a node from G1 , insert a node into G1 , delete an edge from G1 , insert an edge into G1 , and substitute an edge in G1 . This section summarizes how a virtual experiment environment can be used as a platform to train students’ science process skills; how we make computer generate virtual experiment automatically; and, how two graphs can be compared. Before we design the graph-based science process skill diagnosis methodology, Section 3 first reveals the virtual experiment environment and introduces how the student’s science process skills are recorded for further diagnosis.

3 Virtual Experiment Environment We have designed a Story-based Context-aware Knowledge Structure and a revised seven stages of problem solving to generate the virtual experiment automatically. The virtual experiment environment is a web-based application. Both students and teachers use web browser such as Microsoft Internet Explorer and Firefox to access the application. Teachers can build their own knowledge structure for specific science course and access their students’ learning progress. Students will be asked to do virtual experiment in the virtual experiment environment. In the rest of this section, we are going to use a real scenario to explain how the virtual experiment environment works and how the environment trains the student’s science process skills. Ms. Cool is a Physics teacher and wants to train her students’ science process skills in Dynamics. She leads the students to the computer lab and

310

M.-X. Fan et al.

asks them to do Dynamics relevant virtual experiments in the virtual experiment environment. Alex is a student in Ms. Cool’s Physics class. When he gets into the virtual experiment environment, the system first shows him a story book and asks him to choose the story described in the book he interests. Alex chooses the cliff story that he thinks it should be fun. After Alex chose the story, the system asks him to think about the cause and the effect of the hypothesis for the story-based problem as Fig. 2 shows, i.e. formulating hypotheses skill, one of science process skill in Fig. 1. He chooses “to make people’s mass increase will let the stone’s height increase” as the cause and the effect of the hypothesis.

Fig. 2. The cause and the effect of the hypothesis for the story-based problem

With the cause and the effect of the hypothesis Alex have chosen, he then needs to choose the relevant animation to the story-based problem, i.e. classify skill. When Alex makes his choice at this stage, he will be asked to watch different animation clips, and he finally picks the bullet shoot to block animation up. After he made choice and picked-up an animation he thought that it may be relevant to the story-based problem, the system asks him to do object mapping between the problem and the animation, i.e. classify skill. Alex maps the people to the bullet and the stone to the block, under such circumstance, the stone and the people are the story objects and the bullet and the block are the animation objects. After Alex figured out the object mappings, he needs to pick-up the Physics quantity that he should observe in the following experiment in order to solve the story-based problem for each object, i.e. variable identifying skill. Alex thinks the bullet’s mass and velocity and the block’s mass, velocity, and height are important Physics quantities and worth to observe for solving the problem. After Alex chose all the Physics quantities for corresponding object, the system asks him to categorize these quantities into the manipulated variable, the responding variable and the control variable(s), i.e. variable identifying skill. Alex identifies the bullet’s mass as the manipulated variable, the block’s height as the responding variable, and he thinks that other object quantities belong to the control variables. Alex then watches the animation with controller, i.e. play, rewind, and pause, to observe and to record these Physics quantity values. The system will generate relevant charts with these records and Alex can take a look at the charts and think about the relations

Using Graph Edit Distance to Diagnose Student's Science Process Skill in Physics

311

among different objects’ Physics quantities, i.e. constructing graphs skill and analyzing investigations skill. At last, Alex needs to evaluate his hypothesis to see if the hypothesis and the chosen animation can solve the story-based problem for him, i.e. analyzing investigations skill. He can repeat these stages to solve other story-based problems. This section describes the virtual experiment environment and uses a practical scenario to show how the virtual experiment environment works in training students’ science process skills. Section 4 talks the design of the graph-based diagnosis methodology, this research uses the methodology to compare the student’s virtual experiment manipulation behavior with the expected one and to give both the teacher and the student the diagnosis results as feedback.

4 Diagnosis Method In this research, we use the graph edit distance to analyze students’ manipulation behaviors in the virtual experiment environment and diagnose students’ science process skills. For this purpose, a student’s virtual experiment manipulation behavior and the correct one are needed to transform to graphs first. The graph transformation method treats the options at every stage of a virtual experiment as the graph’s nodes, and considers the student’s manipulations and choices as the graph’s edges.

Fig. 3. Student’s manipulation graph of choosing the hypothesis’ effect

Fig. 4. The graph to represent the correct hypothesis’ effect choice

312

M.-X. Fan et al.

For example, Fig. 2 is a virtual experiment stage which asks the student to choose the hypothesis for specific Physics problem. In this example, the student needs to choose the cause and the effect of the hypothesis. The cause and the effect options are physical objects, physical quantity, and the state that this research uses it to present whether the quantity value is going up or down. These options will be transformed to the nodes by the graph transformation method. If the effect a student chose is that the stone’s mass will increase, the graph nodes, i.e. stone (physical object), velocity (physical quantity), and increase (the state), will be connected as Fig. 3 shows. Fig. 4 shows the correct options of the hypothesis. In Fig. 4, the most left node is the root node, is using to distinguish which stage the graph represents. After the student’s graph and the correct graph are transformed, the distance of the two graphs can be used to measure the difference between these two graphs. This research uses the graph edit distance to compute the distance. Different from abovementioned graph edit distance research, this research doesn’t take node costs into consideration due to the two graphs in this research have identical node sets. Also, the cost of substituting an edge does not take into consideration due to all edges are undirected and there is no different meaning attached on the edges, i.e. all edges are same. This research considers only if two nodes are connected. This research defines node level from right to left, which means the rightmost nodes are at level 1. Symbols that this research uses to describe the graph and to design the diagnosis methodology are: (1) uses node(k,i) to represent a node in the graph, where k is the level the node belongs to and i indicates the node is the i-th node; and, (2) uses e{node ( k , j ) , node (( k −1) , j ) } to represent the edge between the node(k,i) and node(k-1,i). The two costs that this research considers are

cost edge_delete (e{node( k , j ) ,node(( k −1) , j ) } ) : the cost of deleting an edge between two nodes from adjacent two levels.

cost edge_insert (e{node( k , j ) ,node(( k −1) , j ) } ) : the cost of inserting an edge between two nodes from adjacent two levels.

□

For instance, the cost of changing Fig. 3 to Fig. 4 will be

cost all = cost edge_delete(e{ stone , mass } ) + cost edge_insert (e{stone , height } ) + cost edge_delete (e{mass , increase } ) + cost edge_insert (e{height , increase } ) , in order to delete the edge between stone and mass and the edge between mass and increase, and to insert the edge between stone and height and the edge between height and increase. In addition to the cost, every node in the graph in this research has level attributes. The left nodes have higher level values, because the left-hand-side nodes and its edges represent the precedent actions and/or choices that the student has made in doing the virtual experiment. So the right-hand-side nodes can’t be connected without its parent nodes, i.e. the left-hand-side nodes, being connected first. For science process skill diagnosis, the higher level a node has, the more important the node is.

Using Graph Edit Distance to Diagnose Student's Science Process Skill in Physics

313

This research thus defines the level weight of a node as weight level ( node( k , j ) ) . The level weight of each node depends on whether the node has connected with other nodes or not in the correct graph. First, all nodes except the root node have same level weight, which is 1 in this research. Second, the children nodes’ weights will be added to its parent node. Taking Fig.5 as example, the level weight of node C is 1 plus the summation of its children nodes’ weights, i.e.

weight level (nodeC ) = weightlevel (node( 2 ,1) ) = 1 + weightlevel (nodeF ) + weightlevel (nodeG ) + weightlevel (nodeH ) = 4.

□

Fig. 5. Level weights of nodes in the graph

Beside the level weight can be used to distinguish the importance of nodes at different levels, the nodes at the same level may have different importance due to its conceptual meanings. For example, in a Dynamics experiment, let’s give the student a hypothesis and three physical quantity concepts: velocity, mass, and modulus of elasticity, and asks the student to choose the effect of the hypothesis. Would it be an issue if the student chooses stone’s modulus of elasticity instead of choosing either stone’s mass or stone’s velocity? The modulus of elasticity of an object actually belongs to another Physics topic rather than Dynamics. If the student chooses another topic’s physical quantity, it might imply that the student may have big misconception and his/her specific science process skill may need to improve significantly. This research defines the conceptual weight for each node weight conceptual (node( k , j ) ) in order to tell the difference between a student’s choice and the correct one when diagnosing the student’s science process skills. The conceptual weights of two nodes at the same level represent how different conceptual meanings the two nodes have. The different concept weights may influence a student’s science process skill diagnosis result. But in our virtual experiment environment, all conceptual weights are considering same due to the system only uses Dynamics relevant concepts which stored in the knowledge structure to generate the virtual experiment.

314

M.-X. Fan et al.

The level weight and the conceptual weight will be used as the coefficient of the graph edit cost:

( weightlevel (node( k-1, j ) )*weightconceptual (node( k-1, j ) ))* (costop (e{ node( k , j ), node( k-1, j ) } )) The costop (enode

( k , i ), node( k +1 , j )

) has three operations:

1. insert an edge, costedge_insert (enode ); ( k , i ) ,node(( k +1) , j ) 2. delete an edge, costedge_delete (enode node ); ( k , i ), (( k +1 ) , j ) 3. and, no change, cost n/a (enode

( k , i ), node(( k +1), j )

The

).

costn/a happens when the connectivity of two nodes is exactly same in both

graphs, which means, there is no necessary to edit the graph. Taking Fig. 5 as example, we suppose the conceptual weights of all nodes are 1. If we want to delete the edge between node A and C, we need:

( weight level ( nodeC ) * weight conceptual (nodeC )) * cost op (e{ A,C } ) = (4 *1) * cost edge_delete (e{ A,C } ) .

□

The total cost (the difference) of two graphs can be summarized as:

∑∑∑ {weight

( l −1) nk nk −1

k =1 i =1 j =1

level

}

(node{k,i} )*weight conceptual (node{k,i} ))*cost op (enode( k , i ) ,node(( k +1 ) , j ) )

where l indicates how many levels the graph has, nk represents how many nodes in level k and nk-1 represents how many nodes in next level. The total cost can represent how different the student’s graph is from the correct one. Furthermore, if the teacher wants to know the student’s specific science process skill level, the maximal cost has to be found. Using the maximal cost, we can understand how the student performed via (1 −

cos t total ( the_stude nt ) ) * 100 % . Max ( cos t total ( possible_ graph j ( the_graph )))

□

where costtotal ( possible_graphj (the_graph) ) means total cost of changing a graph, possible_graphj(the_graph), which has different connectivity combinations, to the correct graph, the_graph. The percentage computed by the diagnosis methodology represents the degree that the student has completed in specific stage, i.e. specific science process skill, therefore the percentage could be represented as the student’s completion degree of specific science process skill. The quantitative approach’s results can then be used to generate the feedbacks to both of the teacher and the student.

5 Experiment Design This research plans to have an experiment to assess the graph-based diagnosis methodology in finding students’ science process skills in this Summer. The participants of

Using Graph Edit Distance to Diagnose Student's Science Process Skill in Physics

315

the incoming experiment will come from either middle school or high school, because our virtual experiment environment currently cover the Physics topics from seventh grade to twelve grade and have hundred Physics experiment animations for generating different Physics virtual experiment automatically. We plan to focus on Dynamics relevant virtual experiments due to these virtual experiments have been approved by Physics teachers and professors. The students who are going to be participating in the incoming experiment are called as the experiment group students. We will collect their genders and past academic achievements in Physics information for further experiment evaluation. The experiment group students will use the virtual experiment environment that we have built and the graph-based science process skill diagnosis methodology proposed in this paper. The incoming experiment requests the teacher to lead the experiment group students to computer lab for using the virtual experiment environment and the science process skill diagnosis system. The incoming experiment will collect the following data for doing quantitative and qualitative data analysis: student’s manipulation behavior logs, computer attitude questionnaire, computer literacy questionnaire, science process skills and its mastery levels (via computerized diagnosis system), genders, past academic achievements in Physics, pre-test results, post-test results, two revised tests of integrated science process skills [3][9] (one will be done before the teacher leads the students to computer lab and another will be done after the students used the virtual experiment environment), and interviews. For assessing if the graph-based diagnosis methodology is useful and understanding the accuracy of the diagnosis results, two ways will be used. First, the researchers will compare the consistency and the agreement degrees among the student’s post-test results, the two revised tests of integrated science process skills, and the science process skill computerized diagnosis results. Second, the Physics teacher will be getting involved in analyzing the student’s manipulation behaviors in virtual experiment environment and grading the student’s corresponding science process skills. The researchers will then compare the agreement degree between the computerized diagnosis results and the human grading results.

6 Future Work This research proposes a graph-based diagnosis methodology to compute the student’s completion degree of specific science process skill. In the following research, we plan to define the student’s achievement level according to the percentage and the virtual experiment difficulty. Similar completion percentages of specific science process skill in two different virtual experiments may have different meanings. For example, Alex had 70% and 50% completion degree of classify skill in both virtual experiment A and B, which doesn’t mean Alex’s classify skill in the two experiments are different if virtual experiment A is an easy one and virtual experiment B is a difficult one. So our future work is to find the method to identify the achievement levels for different virtual experiments and to give both the teacher and the student a more objective feedback about the student’s science process skills.

316

M.-X. Fan et al.

References 1. Assessment of Performance Unit (APU): Science in school. Age 15. Report No. 4 DES, England (1986) 2. Bunke, H.: On a relation between graph edit distance and maximum common subgraph. Pattern Recognition Letters 18, 689–694 (1997) 3. Dillashaw, F.G., Okey, J.R.: Test of integrated process skills for secondary science students. Science Education 64(5), 601–608 (1980) 4. Heh, J.-S., Li, S.-C., Chang, J.-C., Chang, M.: Providing Students Hints and Detecting Mistakes Made by Students in Virtual Experiment Environment. IEEE Transactions on Education 51(1), 61–68 (2008) 5. Kuo, L.-P., Dong, D.-X., Hsu, C.-K., Heh, J.-S.: Design an Enhanced Virtual Experiment Environment Using Science Process Skills on WWW. In: The Proceedings of the 12th AACE World Conference on Educational Multimedia, Hypermedia & Telecommunications (ED-Media 2000), Montreal, Canada, June 25-July 1, p. 1785 (2000) 6. Kuo, R., Chang, M., Dong, D.-X., Heh, J.-S.: Applying Knowledge Map to Intelligent Agents in Problem Solving Systems. In: The Proceedings of the 14th AACE World Conference on Educational Multimedia, Hypermedia & Telecommunications (ED-Media 2002), Denver, Colorado, USA, June 24-29, pp. 1053–1054 (2002) 7. Martin-Villalba, C., Urquia, A., Dormido, S.: Object-Oriented Modelling of VirtualLaboratories for Control Education. Intelligent Systems, Control and Automation: Science and Engineering 38, 103–125 (2009) 8. Mitchell, W.E., Kowalik, T.F.: Creative Problem Solving (1989), http://www.qub.ac.uk/directorates/sgc/learning/Resources/Man agingstress/Filetoupload,119297,en.pdf (retrieved on January 24, 2010) 9. Myers, B.E., Dyer, J.E.: Effects of investigative laboratory instruction on content knowledge and science process skill achievement across learning styles. Journal of Agricultural Education 47(4), 52–63 (2006) 10. Novak, J.D.: Applying learning psychology and philosophy of science to biology teaching. The American Biology Teacher 73, 12–20 (1981) 11. Polya, G.: How to Solve It. Doubleday, Garden City (1957) 12. Sanfeliu, A., Fu, K.-S.: A distance measure between attributed relational graphs for pattern recognition. IEEE Transactions on System, Man, and Cybernetics 13, 353–362 (1983) 13. Tai, K.-C.: The tree-to-tree correction problem. Journal of the Association for Computing Machinery 26(3), 422–433 (2003) 14. Wagner, R.A., Fischer, M.J.: The String-to-String Correction Problem. Journal of the Association for Computing Machinery 21, 168–173 (2001) 15. Wu, S., Chang, A., Chang, M., Liu, T.-C., Heh, J.-S.: Identifying Personalized Contextaware Knowledge Structure for Individual User in Ubiquitous Learning Environment. In: The Proceedings of the 5th International Conference on Wireless, Mobile and Ubiquitous Technologies in Education, Beijing, China, March 23-26, pp. 95–99 (2008) 16. Yan, Y., Liang, Y., Du, X., Saliah-Hassane, H., Ghorbani, A.: Putting labs online with Web services. IEEE Computing 8(2), 27–34 (2006) 17. Fan, M.-X., Kuo, R., Chang, M., Heh, J.-S.: Using Story-based Virtual Experiment to Help Students Building Their Science Process Skills. In: Proceedings of the AACE 22th World Conference on Educational Multimedia, Hypermedia & Telecommunications (ED-MEDIA 2010), Toronto, Canada, June 29-July 2 (2010)

Intelligent Assessment in Math Education for Complete Induction Problems Wolfgang Müller1 and Maren Hiob-Viertler2 1

Media Education and Visualization Group 2 Department of Mathematics University of Education Weingarten, Germany

Abstract. SAiL-M (Semi-automatic Analysis of Individual Learning Processes in Mathematics) is a joint research project with partners from several German universities and financed by the German Federal Ministry of Education and Research, targeted to develop new models to improve the quality of teaching mathematics in early semesters. The major objective of this project is to develop, apply, and evaluate activating learning scenarios and environments for learning mathematics at university level. This includes the development of novel teaching approaches and of novel ways to provide enhanced feedback to student performance on an individual bases utilizing semi-automatic, computerbased assessment tools. Successful approaches identified in the course of the project are collected and documented as best practices in terms of pedagogical design patterns. In this paper, we will provide an overview on talhe SAiL-M project, it's objectives and approaches. We will briefly introduce novel class concepts and teaching approaches being applied in the project. We will present and discuss the Intelligent Assessment paradigm and corresponding tools, which are currently being developed in the SAiL-M project. Here, we will focus on the Saraswati toolset designed and implemented at the University of Education at Weingarten. Finally, we will introduce the concept of pedagogical design patterns and explain, how these are being applied in the project.

1 Introduction Many countries currently observe severe problems to attract students' interest in the fields of math science. As a result, science and industry already suffer from problems of finding qualified graduates to fill their open positions. Many experts agree that this problem can be accounted to a lack of qualified teachers in elementary and secondary classrooms, which again can be attributed to severe deficits and today's teacher education. In mathematics, one major problem is that in traditional education at university level most often the focus rather is on mathematical techniques and standard solutions for given problems than on the promotion of general mathematical processing skills and corresponding competencies. A recent research by TIMSS (Third International Mathematics and Science Study, Bos et al., 2007) provided the unsatisfying proof that in Germany only one child of six depicts elementary mathematical competence after finishing elementary school, X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 317–325, 2010. © Springer-Verlag Berlin Heidelberg 2010

318

W. Müller and M. Hiob-Viertler

and 4% even don't reach this level. Yet, similar problems seem to exist in other countries. Mathematical subjects requiring a larger amount of cognitive processing and transfer skills seem especially problematic. One possibility of reducing those deficits is shown by the OECD in its publication (PISA) „Learners for Life: Student Approaches to Learning“ (see Artelt et al. 2003). It underscores the importance of the individual's own mental frame for learning. Evaluations confirm that students achieve better performances when they are motivated, know effective learning strategies, and have a good self-confidence. Clearly, this gets even more important in teacher training, and in math teacher training in specific. Existing structures in math education at university level regularly amplify the problem. Typical classes are structured into lectures and exercise classes, often reducing math education to theoretical presentations and solving of standardized exercises. Mathematical competences, like problem solving, mathematical reasoning, and communicating, are difficult to mediate in such a scenario. Also, courses in mathematics in early stages of study often have a large number of participants, making it very difficult to provide appropriate individual feedback to students' performances and learning processes. Last not least, math education at university level suffers under similar problems as schools in attracting students' interest for math and science classes, posing a further challenge to math teachers. What is really needed are learning scenarios, where learners can get individual support, especially on how to apply learning strategies effectively and on learning to learn. Also, there is a need for more situations in which self-directed learning can take place. Unfortunately, the two demands – more individual learning and personalized learning processes on the one hand, individual feedback on the other hand – clearly would result in a much higher demand of resources, when introduced into classical learning scenarios. In fact, corresponding learning scenarios only seem possible with the support of appropriate methods of e-learning and suitable tools to allow for personalized learning and individual feedback. In the following, we will provide an overview on the SAiL-M project, which targets to provide solutions for the above stated problems.

2 SAiL-M Objectives SAiL-M (Semi-automatic Analysis of Individual Learning Processes in Mathematics) is a joint research project with partners from several German universities taking up the before-mentioned problems and targeted to develop new models to improve the quality of teaching mathematics in early semesters. This includes the development and evaluation of activating learning scenarios and environments for learning mathematics at university level. An important aspect represents the development of novel ways to provide enhanced feedback to student performance on an individual bases utilizing semi-automatic, computer-based assessment tools. Successful approaches identified in the course of the project are collected and documented as best practices in terms of pedagogical design patterns. In summary, SAiL-M targets to •

Formulate, implement, and publish pedagogical design patterns of activating learning environments for mathematics at university,

Intelligent Assessment in Math Education for Complete Induction Problems

• •

319

Adapt tools for the assessment of learning processes – i.e., tools for the documentation and analysis of processes - and use and evaluate them in learning environments, and Evaluate the effectiveness of process-orientated feedback with various diagnostic methods.

Especially students new at the university shall be introduced self-determined and intentional learning. They shall be given individual support for improving their attitude towards learning. In the following we will discuss approaches taken in the direction on intelligent assessment the field of algebra in some more detail.

3 Intelligent Assessment One approach to increase the level of individual feedback for students is the application of computer-based exercises and assessment tools. Computer Aided Assessment (CAA) refers to a number of approaches to assess students’ performance using a computer. CAA promises that test results may be analyzed and compared with minimal effort in minimal time. These time and resource savings would allow more regular assessments than otherwise possible. As a result, teachers could gain more detailed knowledge of students’ progress and, thus, identify problems earlier. Last not least, tests could be tailored to match students’ abilities. A problem with typical CAA techniques is that they are usually restricted to assess factual knowledge based on objective tests and multiple-choice questions. Furthermore, the construction of these tests is very time consuming and requires very specific knowledge. Intelligent Assessment refers to yet another approach to IT-based assessment (Bescherer et al. 2009). It is based on assessment tools, which not only assess and analyze students’ products, but also the processes based on which these products were generated. In general, the semi-automatic approach means that assessment tools target to detect and filter standard solutions and standard errors. Unusual and novel solutions, which cannot be categorized automatically, are forwarded to the teacher for ‘human’ assessment. Thus, a teacher may focus on exceptional solutions providing interesting aspects for classroom discussion, and there is no need to read ‘uninteresting’ standard solutions. A problem with analyzing processes for feedback or assessment is the sheer number of possibly correct and incorrect solutions. In a lot of training programs for mathematics it is still quite common that certain correct answers obtained by alternate solution strategies are evaluated as incorrect. Furthermore mistakes are of an individual nature and in their possible number unlimited. As a consequence, intelligent assessment must not restrict students to the preconceived solution paths of test developers. Instead, all different – and correct – solutions must be allowed. In fact, this requires the integration of expert systems to minimize the number of possibly unrecognized solutions, as these would have to be assessed by a teacher manually. Saraswati (Bescherer, Müller, Heinrich and Mettenheimer 2004) represents an example for an intelligent assessment framework in the field of algebra, which is being developed and extended in the context of the SAiL-M project. The Saraswati system provides a complete framework for authoring and solving corresponding exercises as

320

W. Müller and M. Hiob-Viertler

Fig. 1. Web-service architecture of the Saraswati system

well as their assessment and analysis. The Saraswati framework is based on several components, most important test components allowing students to interactively develop and edit solutions to algebraic problems, a web-service wrapping an analyzer component for assessing solutions and solutions processes, as well as an authoring component for teachers to develop adequate exercises, a logging backend and a statistics front end to collect data on student solutions and to analyze solutions, both on individual and class level. The analyzer component integrated in the web-service represents the central element of the Saraswati system. It allows the analysis of student's, which includes the evaluation of solutions, the detection of errors, the classification of detected errors based on appropriate heuristics, and the reporting and logging of errors. For this, Saraswati utilizes CAS (computer algebra system) functionalities provided by the CAS Maxima (Maxima 2009) and leveraged via a web-service (see Figure 1). The CAS is mainly being used to provide the necessary functionality to assess the correctness of mathematical rewritings in the process of solving a specific exercise. In addition, it is utilized to access individual mathematical terms in the process of error classification, as explained in more detail later. The main assessment logic is mostly being represented in terms of specific services provided by the Saraswati web service. Saraswati utilizes MathML (W3C 2010) to represent mathematical terms and expressions. That is, Saraswati client components providing the user interface for students to solve mathematical exercises and to enter their solutions, are expected to produce MathML expressions, bundled in an XML-based exercise set. These exercise sets are specific to the mathematical problem domain. For instance, for linear systems of equations (LSEs) it contains of the original exercise and a list of LSEs, representing a student's complete solution for this exercise. MathML based solutions are being passed via the Saraswati web service to the CAS for evaluation. Also, MathML is being utilized to represent mathematical expressions in the context of feedback to students and logging for teachers. Figure 2 gives a more detailed view on the Saraswati architecture. The analyzer component plays the key role in the Saraswati concept. It targets to grasp the process of finding a solution, not only the correctness of the final result. For this, the correctness of each individual rewriting step is validated. In case of errors, heuristics are being applied to identify the type of error and possible corrections.

Intelligent Assessment in Math Education for Complete Induction Problems

321

Fig. 2. Complete view on the Saraswati architecture

These heuristics are usually based on data on typical student errors, and the implementation of corresponding detectors. Following rewritings are further examined under consideration of previous errors. Clearly, the specific set of error detectors can hardly cover all different types of errors. Instead, the before-mentioned semiautomatic approach is taken, and errors that could not been further classified are automatically being forwarded to a teacher or tutor for human assessment. All assessment results are collected and compiled to provide statistics and detailed information on both individual students and class level. We currently apply the Saraswati approach in two domains: solving linear systems of equations (LSE) and complete induction. Details on the Saraswati system for LSEs can be found in Bescherer et al. 2004. In the following, we will explain in some more detail the concepts applied in the field on mathematical induction. Mathematical induction has been selected as an application field for example due to the fact that most students find it difficult and often fail to get fundamental understanding of this proof concept. Moreover, a characteristic of this proof concept is the standardized procedure, almost comparable to an algorithm, to perform this type of proof. Assessing student solutions for mathematical induction proofs therefore requires not only a final evaluation of the correctness of a proof, but a more detailed analysis and evaluation of the applied procedure and approaches to differentiate between simple arithmetical mistakes and more severe problems based on some misunderstanding of proof concept. Mathematical induction therefore represents a perfect application field for the Saraswati concepts. As mentioned above, it is clearly impossible to provide adequate assessment and feedback on mathematical exercises on a general bases. In the Saraswati concept, specific detectors are being applied to identify errors, and problem-domain specific heuristics are used for error classification. For this, the problem domain has to be

322

W. Müller and M. Hiob-Viertler

restricted. In the context of mathematical induction we therefore currently concentrate and restrict our current developments on those kind of proofs including sum formulas. For instructional reasons, we structure exercises in the area of mathematical induction in three steps, corresponding to the three major steps in mathematical induction proofs: The based case, showing that the equation to proof holds for some initial value of a variable (usually n = 0 or n = 1), • •

The inductive hypothesis, representing the assumption that the given equation holds for a specific value (e.g., n = k), and The inductive step, showing that given the assumption the equation also holds for a natural increment of the value assumed in the inductive hypothesis (e.g., k + 1).

For the base case of a mathematical induction proof, we expect students to •

Identify the appropriate variable of an equation, over which the mathematical induction has to be performed, • Select an appropriate start value for this variable, • Substitute the variable with this start value accordingly, and • Perform an adequate simplification to show that the equation holds for this base case. For the inductive hypothesis, we expect students to deliberately select some value (e.g., k), for which they assume the equation to hold. For the inductive step, students shall • • •

Substitute the induction variable with some natural increment of the value, for which they assumed the equation to hold (e.g., k + 1), Perform adequate substitutions to allow a referral to the inductive hypothesis, and Simplify the equation to a state where the correctness of the equation can clearly be seen.

For this, we structured mathematical induction exercises in three corresponding subparts. In the first part, students have to provide a proof on the base case. Here, they have to enter both, the name of the variable over which they perform the induction, and the corresponding base value. Next, they have to provide a sequence of rewritings of both, the left and right hand side of the equation with the variable substituted with the base value. An assessment of this part of the solution by the Saraswati analyzer component can then later be performed based on whether the variable name and base value match the ones specified by the designer of the exercise. Furthermore, the analyzer checks whether the substitution has been performed correctly, and whether the provided rewriting of the equation contains any arithmetic errors. This is done based on a comparison of the left hand and right hand terms of subsequent rewritings respectively. Here, in case of the detection of an error some specialized error detectors are activated. Specific errors that is being tested for, are a wrong or an incomplete substitution with the base value. Solving the second part of an exercise does in fact only require a simple rewriting of the equation, for instance exchanging a variable n by a value k, for which it is

Intelligent Assessment in Math Education for Complete Induction Problems

323

Fig. 3. First part of the corresponding interface

being assumed that the equation holds. The current version of our assessment tool indeed requires a corresponding rewriting of the equation. In future version, however, we plan to take this burdon of manual rewriting of the equation from the student. Instead, we intend to map this part of the exercise to an adequate multiple choice test, including appropriate disctractors to make this task a bit more difficult. For the inductive step, finally, students will have to provide the initial substitution (e.g., k ® k + 1) and again a sequence of rewritings of the corresponding equation to a final form, clearly depicting the equivalence of the left and right hand term. Assessment of this part of a solution in the analyzer component for correct substitution and arithmetic errors is performed similar to step 1. In addition, the final result of the rewritings is further analyzed to identify whether a form has been reached where the equivalence of the left and right hand term can easily be identified. In fact, this part of the assessment proofs to be the most difficult one. At this point, we utilize our before-mentioned restriction on mathematical induction proofs containing sum elements. This allows us provide specialized detectors that look for specific sum and fractional elements in the equation. In specific, we request students to utilize the inductive hypothesis in the rewritings at one point. A specific detector checks whether the inductive hypothesis appears in the rewritings. Combined, these detectors allow to detect not only specific errors (e.g., error in substitution, arithmetic error), but also to assess the progress a student could make in developing a proof (e.g., correct substitution, utilization of inductive hypothesis, completion of the inductive step).

Fig. 4. The inductive step

324

W. Müller and M. Hiob-Viertler

Figure 3-4 depicts a partial view of the corresponding interface of the Saraswati client for students to develop solutions online. In the current version, mathematical expressions such as sums and exponents have to be entered in terms of valid Maxima expressions. However, these expressions are then rendered in standard mathematical form utilizing MathML rendering. This is being achieved by rendering mathematical expressions into MathML using corresponding functionalities of the Maxima-based web service and AJAX-based client functionalities. At the same time, this provides for a syntactical checking of mathematical expressions. A first prototype version of the Saraswati framework for mathematical induction has been realized. Currently, first user tests are being performed to increase the systems's usability and to identify problems in the workflow.

4 Summary and Conclusion In this paper we provided an overview on the SaiL-M project, it's objectives and first results. We introduced new teaching concepts being applied in the project at university level, which focus on a higher degree of self-determined learning and activating learning structures. E-learning tools based on the concept of Intelligent Assessment play an important role in the SAiL-M philosophy. We explained this at the example of Saraswati, a specific intelligent assessment tool for the field of algebra being developed at the University of Education Weingarten. We discussed the Saraswati concept and framework in some detail. We also discussed the approach taken in the SAiL-M to document successful applications of such tools in terms of educational design patterns. Finally, we discussed briefly the concept for evaluation being applied in the project. The SAiL-M project represents an elaborate effort to enhance to quality of math education, especially in math teacher education, in the first semesters. In this project, a number of methods and tools following the idea of Intelligent Assessment have been developed in the last months. First small-scale evaluations provided positive results. We expect that the Intelligent Tools, such as Saraswati, will allow us to develop new forms of class structures at university level in the future, allowing for more selfdirected learning and for a much higher degree of activating learning scenarios.

Acknowledgements The SAIL-M project ist funded by the German Federal Ministry of Education and Research (BMBF) in the program Empirical Educational Research. We would like to thank the SAIL-M project team for input, support, and valuable feedback. We would like to stress that considerable parts of this paper are based on collaborative results from the SAiL-M project. We also would like to thank Felix Tscheulin for his support in the development of the Saraswati analyzer web-service components.

Intelligent Assessment in Math Education for Complete Induction Problems

325

References 1. Artelt, C., Baumert, J., Julius-McElvany, N., Peschar, J.: Learners for Life: Student Approaches to Learning. In: Results from PISA 2000, OECD, Paris (2003) 2. Bescherer, C., Kortenkamp, U., Müller, W., Spannagel, C.: Intelligent Computer-Aided Assessment in Mathematics Classrooms. In: McDougall, A., Murnane, J., Jones, A., Reynolds, N. (eds.) Researching IT in Education: Theory, Practice and Future Directions, pp. 200–205. Routledge, New York (2009) 3. Bescherer, C., Müller, W., Heinrich, F., Mettenheimer, S.: Assessment and Semi-Automatic Analysis of Test Results in Mathematical Education. In: Cantoni, L., McLoughlin, C. (eds.) World Conference on Educational Multimedia, Hypermedia and Telecommunications, vol. (1), pp. S3013–3018. AACE, Norfolk (2004) 4. Bos, W., Bonsen, M., Baumert, J., Prenzel, M., Selter, C., Walther, G. (eds.): TIMSS 2007. Mathematische und naturwissenschaftliche Kompetenzen von Grundschulkindern in Deutschland im internationalen Vergleich, Waxmann, Münster (2008) (in German) 5. Maxima, Maxima, a Computer Algebra System (2009) (last visited: February 20, 2010), http://maxima.sourceforge.net/ 6. National Council of Teachers of Mathematics (NCTM), Principles and Standards for School Mathematics (2000) (last visited: February 20, 2010), http://standards.nctm.org/ 7. W3C, W3C Math Home (2010) (last visited: February 20, 2010), http://www.w3.org/Math/

Research on the Method of Recomposing Learning Objects and Tools in Adaptive Learning Platform Pan Xie, Longmei Ye, Yueming Huang, Youwei Chen, and Liwu Lin Audio-Visual Education Center, Wen zhou, Zhe Jiang, China [email protected], [email protected], {735396408,78389752,3993953}@qq.com

Abstract. At present a large number of learning platforms are the accumulation of educational resources. In the face of chaotic educational resources, learners become overwhelmed and unable to find their own resources effectively. Adaptive learning platform aims to build learning environment and provide learning objects based on learner’s individualized information usage behavior, habits, preferences and etc. It can guide learner’s learning activities as a teacher, and enhance learning efficiency. This paper presents an ontology and sequence mining based selection method for learning objects and learning tools in adaptive learning platform. It mines the learning scheme through the sequence mining method to achieve personalized learning scheme. Besides, according to the learning scheme, it integrates the existing e-learning tools to build learning environment by service composition method, and aggregation all the related learning objects into learning content by using ontology technology to describe the semantic relationship between learning objects. Keywords: Learning Object Ontology, Sequence Mining, Service Composition, Adaptive Learning Platform.

1 Introduction At present, a large number of learning platforms are scattered in the network. The learning environments building by the existing learning platform are mainly focused on the exhibition of the content, practice and testing, as well as the level of coordination and guidance. They generally are lack of concern on the differences between different types of knowledge of the subjects and the learners’ personalized characters. The learning modes in the existing online learning platforms can not meet the needs of different knowledge types and the learners’ personalized requirement. The learning objects could not automatically adjust to the profile of individual learners [1], which results in the duplication study of knowledge, the non-coherent and non-deep of resource studying, and so on. Adaptive learning"[2] refers to that on the basic of grasping the learning rule, the learners can take the initiative to choose the learning method according to learning content, learning scenarios, learning conditions, personal characteristics and factors of learning,, and organize and regulate learning activities. It ultimately manifested as the needs of willing to learn, good at learning, well in the creation, to achieve their own X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 326–336, 2010. © Springer-Verlag Berlin Heidelberg 2010

Research on the Method of Recomposing Learning Objects and Tools

327

development. Some of the elements of adaptive learning include: monitoring student activity, interpreting the results, understanding students' requirements and preferences, and using the newly gained information to facilitate the learning process[3].In the usually adaptive learning platform, there are log module, knowledge representation module, resources integrated modules, learning tools modules, on-line testing and online paper module, real-time free discussion area and the thematic discussion forum module, the learning process records, query module. To build an adaptive learning platform [4], the key problems include how to integrate different learning tool and learning objects efficiently. In this paper, we proposed an ontology and sequence mining based selection method of learning objects and learning tools. First, we mine the learning scheme through the sequence mining method to achieve personalized learning scheme. Then, according to the learning scheme, we integrate the existing e-learning tools to build learning environment by service composition method, and aggregation all the related learning objects into learning content by using ontology technology to describe the semantic relationship between learning objects. It can improve the existing learning model, improve targeting, save the learning time, increase learning efficiency and achieve the best learning results. Section 2 presents the architecture design of the method of recomposing Learning objects and tools in adaptive learning platform. Section 3 introduces learning scheme decision making based on sequence mining. Section 4 introduces the ontology based description mechanism of learning objects and learning tools. Section 5 presents the service composition and learning object aggregation method based on ontology. Section 6 summaries the main idea of the paper.

2 The Method of Recomposing Learning Objects and Tools in Adaptive Learning Platform In order to meet the individual learning needs of learners, this paper give the method of recomposing learning objects and tools in adaptive learning platform. We mines the learning scheme suitable for the individual learner based on sequence mining algorithm, Then, according to the learning activities of the selected learning mode, we build of the personalized learning environment through services composition, and provide learning content though learning object aggregation. The services composition aims to generate a sequence of software functions by matching and integrating existing learning tools that are encapsulated in web services. The recomposing of learning objects is to provide suitable learning objects by describing and matching learning objects based on semantic ontology. As shown in Fig. 1, the architecture of our method includes five engines, which are Sequence Mining Engine, Service Composition Engine, Learning Object Composition Engine, Service Matching Engine and Learning Object Matching Engine. Sequence Mining Engine is used to mine rules from the learners’ using record and learning experience library in order to find suitable personal learning scheme to the learners. Service Composition Engine: its main function is composing the tools as services to form the personalized learning environment according to the learning scheme. First, it analyzes the learning activities in the learning scheme according to the requirement of

328

P. Xie et al.

the learning mode and forms the set of the software supporting tools which support the leaning mode. Then it uses the service matching engine to find the services and return the service composition sequence. This service sequence is the personalized learning environment. Service Composition Engine can generate the service sequences which meet the learning mode by the description of the requested service and the description of the services in the service library, and return the composition scheme to the Learner. Learning Object Composition Engine can aggregate or compose the individual learning objects into one learning object set on the basis of the knowledge points and the relationship between knowledge points according to according to the requirement of the leaning scheme. Fist, it analyzes the learning objects needed according to the requirement of the leaning scheme and the learning content. Then it uses the learning object matching engine to find the learning objects from the learning object library and compose the learning objects into the learning object sequence which is suitable for the learning of the knowledge point. Service Matching Engine is used to carry out service matching according to learning modes. Ontology Library stores domain ontology which is used to describe semantic information of services and support semantic matching of services. Service Library stores all the published domain services which are described by expanded OWL-S. It can register the semantic information of the services and provide semantic matching for the services [5]. Moreover, it can organize and manage the services by categories. Matching Module uses OWL Reasoning Machine to do service matching on function and non-functional description based domain ontology library and service library. Learning object Matching Engine can match the knowledge points with learning objects according to relationship of knowledge points and the requirement of the learning scheme the knowledge points of the learning content. It can consider the learning context environment and find the most suitable learning objects. Learner Learning Content BPEL-WS

Learning Objects

Learning Modes Learning Object Composition Engine Service Composition Engine Learning Modes Sequence Mining Engine Learning Object Matching Engine

Service Matching Engine

Service Library

Ontology Library

Matching Module

OWL Reasoning Machine

Service Using Record

Learning Experience Library

Learning Object Library

Matching Module

Object Ontology Library

OWL Reasoning Machine

Fig. 1. Ontology and sequence mining based adaptive learning model

Service Library stores the existing E-learning software and tools. The software and tools can be wrapped into services. Learning Object Library stores all the learning objects for all the knowledge. In the realization of the platform, the service matching

Research on the Method of Recomposing Learning Objects and Tools

329

engine and the learning object matching engine can be combined together. They can find the learning objects and learning tools suitable for the leaning activities according to the requirement of the learning scheme. Learning Experience Library is the set of the learning scheme which is summarized from the learning of some kind of knowledge. Under this model, the specific process of the adaptive learning platform based on ontology and data mining are as following. (1) Build the learning object library and service library; (2) Collect the learners’ using information of the resource platform and store the information into the service using record library; (3) Summary and conclude different learning modes for different knowledge, translate the modes into inference rules, and store the learning mode into learning experience library; (4) According to the sequence mining algorithm use the Sequence Mining Engine to mine the personalized learning modes for the learner according to the type of the learning knowledge; (5) Use the Service Composition Engine to analyze the learning activities in the learning scheme according to the learning mode and form the set of the software supporting tools which support the leaning mode; (6) Use the Service Matching Engine to find the services and return the service composition sequence which supports the adaptive learning environment; (7) Use the Learning Object Composition Engine to form the set of the learning objects corresponding to the learning knowledge points in the learning content for the learner according to the requirement of the learning mode; (8) Uses the Learning Object Matching Engine to find the learning objects from the learning object library and compose the learning objects into the learning object sequence; (9) Import the learning object sequence into the learning environment; (10) Track the learner's learning behavior and conclude more learning modes according to study results to enrich the learning experience library.

3 Learning Scheme Decision Making Based on Sequence Mining According to the learning theory, take the nature of the learning activities (acceptance-explore) and the learning organization forms as the analytical standards of the learning mode, we can sum up ten main modes, such as case-based learning mode, WebQuest-based learning mode , the concept map-based learning mode, multiple intelligence-based personalized learning mode, resources-based project learning mode, web-based collaborative learning mode, project-based learning mode, ePortfolio-based learning mode, problem-based learning mode , context-based learning mode[6]. However, there are differences between the learning mode and strategies of different subject contents. And there are differences between the characteristics and rules of different learners. Therefore, for different subject knowledge points and different learners, there are different learning modes. In order to achieve adaptive learning, we first need constitute the suitable learning scheme for the learning content for different learners. To solve this problem, we propose a sequence of mining-based learning scheme decision making method. This method has three steps.

330

P. Xie et al.

First, classify the contents of the subject knowledge into three types. They are declarative knowledge, procedural /methodological knowledge and strategies / controlling knowledge [7]. (1) Declarative knowledge is the species of knowledge that is, by its very nature, expressed in declarative sentences or indicative propositions. For example knowing that "A cathode ray tube is used to project a picture in most televisions" is declarative knowledge. Declarative knowledge is assertion-oriented. It describes objects and events by specifying the properties which characterize them; it does not pay attention to the actions needed to obtain a result, but only on its properties. (2) Procedural / methodological knowledge is different from other kinds of knowledge, such as declarative knowledge, in that it can be directly applied to a task. For instance, the procedural knowledge one uses to solve problems differs from the declarative knowledge one possesses about problem solving. (3) Strategies / controlling knowledge: control the behavior of the rule engine Secondly, make use of the mining technology to get the learning scheme for group of learners and individuals on learning different kinds of knowledge. The mining method mainly solves two problems. First, for the group of learners, through mining we can sum up the learning process for the knowledge types or specific knowledge points, get the sequence of the learning activities and form the common rules which are learning scheme. For example, for the concept knowledge, the students mainly use the concept map-based learning mode and problem-based learning mode. Secondly, for some particular learner, mine the history data when he learns certain kind of knowledge which is one of the three kinds of knowledge and form the learning scheme on certain kind of knowledge for his preference. Thirdly, through synthesizing the common learning mode on certain kind of knowledge and the learning mode of personal preferences and referencing the ten kinds of learning modes which are summed up based on the learning theory, constitute the specific learning scheme for further study of the learners. The key point of realizing this method is the mining of the learning mode. We adapt the sequence mining technology [14] to mine the learning history data of the learner, and get the learning scheme through the sequence of learning process. The specific realization algorithm mainly includes two steps. The algorithm first finds all the modes whose support rate is more than given threshold. Then it finds the maximum from the modes as the motif. Algorithm 1: The mining algorithm of finding all the basic learning processes whose support rate is 2 Suppose database D stored all the learning history data of the learners. For learning group or one learner, mine the common learning process by the method below. Step1: [Find U(T)]// Find all binary service sequences in the service composition tree T Step2: [Find M(D,2,1)] // Find the mode whose length is 1 and support rate is 2 from sequence database D. Step3: [Find M(D,2,2)] //Find all the modes whose length is2 and support rate is larger or equal to 2

Research on the Method of Recomposing Learning Objects and Tools

331

Step4: [Find M(D,2,l)] // Find all the modes whose length is 1 and support rate is larger or equal to 2 Step5: M(D,2)=M(D,2,1) M(D,2,2) … M(D,2,l)

∪

∪ ∪

Algorithm 2: Find the mode whose support rate is larger than the given threshold Step1: [Find M(D,2)] Call algorithm1 Step2: [Find M(D,h)] The sequence we need is the maximum sequence in M(D,R), R is the support rate. We can set the value of R to control the mining depth.

4 Ontology Based Description Mechanism of Leaning Objects and Leaning Tools In adaptive learning, after identifying the learning scheme for the learner, we need build the learning environment and provide the learning content for him according to the knowledge content which he wants to learn and his learning mode. The learning environment is created through service composition and service-oriented integration of learning tools. Moreover, the learning content is the restructure of the learning objects which are going to be learned. Whether the integration of the learning tools or the aggregation of learning objects, we must first resolve their description problem, and the descriptions should include semantic information. Therefore, we propose ontology-based learning objects and learning tools description method. 4.1 Ontology Description Method of Learning Objects There are many representation methods of learning objects [8]. They describe the learning objects form three levels, object level, concept level and representation level [9]. However, they can not describe the semantic information of the learning objects and the relationship of learning objects. Ontology is a "formal, explicit specification of a shared conceptualization"[10]. It provides a shared vocabulary, which can be used to model a domain – that is, the type of objects and concepts that exist, and their properties and relations Domain ontology models a specific domain, or part of the world. It represents the particular meanings of terms as they apply to that domain. It enables machines to communicate with each other on public knowledge library, and consequently, realizes interoperation between heterogeneous applications. Ontology description method of the learning objects is to descript the attributes of the learning objects by e-learning domain ontology. In this paper, we build learning object ontology to describe learning objects by extending OWL [11] according to IEEE Learning Technology Standards Committee (IEEE LTSC)’s Learning Object Metadata [12]. IEEE LOM which specifies the syntax and semantics of Learning Object Metadata, IEEE 1484.2: Learner Model which specifies both the syntax and semantics of a 'Learner Model' which will characterize a learner and his or her knowledge/abilities, Learning object domain ontology is the basic concepts and relationship between concepts included in the domain of the leaning objects. It is used to establish dynamic relationship and share the information. The knowledge points of the learning objects have several basic relationships.

332

P. Xie et al.

(1) Upper – lower relation: it is used to present the relation between conceptual knowledge. The upper concepts have more extensive extension than lower concepts. For example, “flower” is the upper concept of “mum” .This relation has transitivity. (2) Part-whole relation: It can be used to all kinds of knowledge. It is used to present one knowledge is part of another knowledge. For example, complex knowledge includes simple knowledge. This relation has transitivity. (3) Forward-subsequence relation: It can be used to all kinds of knowledge. It is used to present one knowledge is the background of another knowledge. This relation has transitivity. (4) Equivalent relation: It can be used to all kinds of knowledge. It is used to present the contents which are related by one knowledge and other knowledge are the same. This relation has transitivity and symmetry. (5) Correlativity: It can be used to all kinds of knowledge. It is used to present that there are some relation between one knowledge and another knowledge, but they are not equal entirety and don’t have the forward subsequence on logic. It is usually used as the reference link of the learning. As the inner relation and equal relation in OWL can not learning object totally, we build the extended learning object ontology according the type of the knowledge and subject by reference to the classification on the knowledge in cognize psychology .Beside, we also consider the attribute of the learning tools and learning activities which the leaning object are suitable to. 4.2 Ontology Description Method of Learning Tools In order to realize the interoperation between heterogeneous learning tools, we wrap the tools into web service and descript them. There are several description methods of web service. They are WSDL and OWL-S. WSDL only supports the syntax description of service interface information, does not support semantic description .In order to solve the semantic heterogeneous problem of services, it is necessary to add semantic labels to the service. The former adds semantic information to service interface description. But it only provides parts of the semantic description, so it can not support automatic service composition very well. The later describes the properties and function of web services based on OWL, which has become widely accepted and used as a standard .This paper also used OWL-S to describe the service in educational software.OWL-S [13] is an ontology within the OWL-based framework of the Semantic Web for describing Semantic Web Services. It is an emerging standard to add semantics and an upper ontology for describing properties & capabilities of web services using OWL.

5 Ontology Based Service Composition and Leaning Object Aggregation Method On the basic of the ontology description of the services and learning tools, the key of building learning environment is the composition method of the services. At the same time, despite the learning objects is not a software tool, and its semantic description is also different with learning tools. But in broad sense, the digital learning objects can

Research on the Method of Recomposing Learning Objects and Tools

333

also be viewed as a kind of software, a special kind of service. Its aggregation method can be achieved by adapting the same mechanism with learning tools. Just at the time of service matching, they adapt the different standards. Therefore, we propose an ontology-based service composition method which is suitable for the learning object aggregation and the learning environment building. Since the learning scheme is a set of different learning activities, and each activities needs certain learning objects and e-learning tools to support. We can use service composition method to compose the individual learning tool and object together according to the input and output information of the learning scheme. Our service method includes four steps. First, judge composition possibility of services with services’ semantic correlation degree, and construct the parameter-service figure. Second, build And/or tree of service composition by merging parameter node of parameter-service figure. Third, extend the service composition tree through mining the functional relationship between the services which have parameters and the services which have no parameters. Last, cut And/or tree of service composition to obtain optimal service composition sequence. 5.1 The Judgment of the Services Composition Possibility and the Construction Parameter-Service Figure To compose services, the first thing is to judge the composition possibility of the requested service. The main idea of our method is to find a collection of services starting from the user’s input parameters and go forward by chaining services which introduce new intermediate variables until they deliver the user’s expected output parameters. If there is a collection of services which match all expected output parameters provided by the requested service, the requested service can be composed. And the result is shown as the parameter-service figure. Algorithm 3: the building of the parameter service figurer For the service request WSR (Category, I (WSR), O (WSR)) proposed by the user, Category is the kind of the service WSR, I (WSR) is the set of the input parameters of WSR. O(WSR) is the set of the output parameters of WSR. Input: WSR , S //S is the set of services Output: ParameterServiceFigure Step1: I = I(WSR) //I (WSR) extracts the set of input parameters of the service request Step2: for each service Si S{ if( Category(Si) = Category(S) & I(Si) ⊆ I & O(Si) ⊄ I(WSR) ){ put Si in SM // mark Si as the intermediate service and put it in the set of the intermediate services SM put parameters in O(WSi) - I(WSR) in M // Store the new intermediate parameter varibles in M if ( I M ⊆ O(WSR) ){ WSR can be composed by services in SM return ParameterServiceFigure }}} Step3: while(S-SM != Φ ){ for each Si’ S-SM {

∈

∪

∪

∈

334

P. Xie et al.

∪

∪

if(I(Si’) ⊆ I M & I (Si’) ∩ M != Φ & O(Si’) ⊄ I M){ put Si’in SM I = I U (U I(WS’j)) M = UO(WS’j) - ( UO(WS’j) ∩（ I M） ) if(I M ⊆ O(WSR)){ WSR can be composed by services in SM return ParameterServiceFigure}}}else return 0;// WSR is uncombinable. Fig.2 shows parameter service diagram which is organized by the parameter nodes and services in the service composition field.I (WSR) is {P1, P2, P3, P4, P5}, O(WSR)is {P9, P10}.

∪

∪

Fig. 2. Parameter service figure

5.2 And/or Tree of Service Composition s And/or tree of Service Composition describes the relationships of services which match all expected output parameters provided by the requested service. Its edges are weighted by the weight average of service correlation degree and service quality. By merging the parameter nodes in the parameter-service figure according into service nodes, we get a And/or tree of services. The merging algorithm is shown as bellow: For every parameter node in the figure: (1) For the parameter nodes whose in-degree is 0, delete these nodes and the edges whose start points are them. (2) For the parameter nodes whose out-degree is 0, if these nodes are not belong to the output collection of the requested service, then delete these nodes and the edges whose endpoints are them. (3) Delete free nodes whose in-degree and out-degree is 0. (4) For the nodes whose out-degree is 0, if these nodes belong to the output set of the requested service, then first delete those nodes. Then add a new service node to replace the output collection of the requested service and make the edges which point to these nodes point to the new Service Node SF. (5) Make all output service nodes point to SF. (6) For the parameter node whose in-degree is 1, delete the node n1, the edge e1 whose endpoint is n1 and the edge e2 whose start point is n1. At the same time add the new edge whose start point is e1 and endpoint is e2.If n1 belongs to the output set of the requested service, then add the new edge whose start point is e1 and endpoint is SF. (7)For the parameter node n2 whose in-degree is more than 1,replace n2 as service node s2 .Make the former edges whose endpoint is n2 point to s2 and the relationship of

Research on the Method of Recomposing Learning Objects and Tools

335

these edges is “or”. If n2 belongs to the output collection of the requested service, then s2 is marked as “output service node”. (8) In addition to the edges of "or" relationship, other edges which point to the same node have the relation of “and”. (9) Finally, we get the And/or tree which shows the relationship between services, as shown in Fig. 3.

Fig .3. And /or tree of service composition

5.3 Extend the Service Composition Tree According to Sequence Mining The and/or tree of service composition above gets the execution relationships of services according to the parameters of the services. But it only considered the sequential execution of services and didn’t analyze the functional relationships of services .Besides, it is not suitable to the services of information provided which didn’t have the input and output parameters. The tree can not reflect the user's personality characteristics. So in the step we mine the services without parameters which associated with the services in the tree and add these services to the tree to extend the tree.The realization steps are as follows: By using a linear mining algorithm [14], first we analyze the user's service usage records. Then find the mode, whose support rate is greater than a given threshold. At last, extend the tree accord to the mode. The expanded figure of service composition tree is as shown in Fig.4:

Fig. 4. The service composition tree extended

5.4 The Method of Optimizing Service Composition Sequence In the And/or tree above, the root node SF is the requested service and the other nodes are the composition sequence of SF. If all the sub-nodes of one node n have the relation “and”, then n only can be executed after all the sub-nodes finish. If all the sub-nodes of one node n have the relation “or”, then n can be executed after any sub-node finish. To obtain the optimal composition sequence of services which can meets the requested

336

P. Xie et al.

conditions, it is necessary to cut the And/or tree above and traversal by level. The main idea to do this kind of cutting is cutting the edges of And/or tree according to the weight. For Fig.4, the composition sequence for SF is S1, S1’, S2’, S3, S4, S6.

6 Conclusion This paper studies the method of recomposing the learning objects and tools in the adaptive platform and provides ontology and sequence mining based adaptive learning model. This method can build the adaptive learning supporting environment by composing the service tools according to the learning mode. At the same time, we can organize the learning objects according to the learning content based on ontology. This method can reuse the existing resource and resolve the interoperation problem between different resources library. The future work is to establish the learning model library.

References 1. Prentzas, J., Hatzilygeroudis, I., Garofalakis, J.: A Web-Based Intelligent Tutoring System Using Hybrid Rules as Its Representational Basis. In: Cerri, S.A., Gouardéres, G., Paraguaçu, F. (eds.) ITS 2002. LNCS, vol. 2363, p. 119. Springer, Heidelberg (2002) 2. Moore, M.G.: Toward a Theory of Independent Learning and Teaching. J. The Journal of Higher Education 44(9), 661–679 (1973) 3. Paramythis, A., Loidl-Reisinger, S.: Adaptive Learning Environments and eLearning Standards. Electronic Journal of eLearning 2 (2004) 4. Brusilvovsky, P., Paylo, C.: Adaptive and Intelligent Web-based Educational Systems. J. International Journal of Artificial Intelligence and Education 13, 156–169 5. Zhang, Y., Wang, F., Zhang, R.: Semantic Matching Method Based on Ontology for Grid Services. J. Computer Engineering, 4 (2007) 6. Ying, H., Lihui, P., Xinde, Y.: The Role of Knowledge Classification in the Instructional Design. J. Educational Review (2008) 7. Zhixian, Z.: Information teaching model - Theory and Practice. C. Education, Science Press (2005) 8. Milošević, D., Brković, M.: Adaptive Learning by Using SCOs Metadata. J. Interdisciplinary Journal of Knowledge and Learning Objects 3 (2007) 9. Yu, P.T., Li, H.W.: Chia Ming Liu. J. The Study of Adaptive Learning Sequence in the Knowledge Space based on Formal Concept Analysis, NCS (2005) 10. Gruber, T.: A translation approach to portable ontology specifications. In: Knowledge Acquisition, pp. 199–220 (1993) 11. Web Ontology Language (OWL) (2004), http://www.w3.org/2004/OWL 12. IEEE LTSC, http://ieeeltsc.org/ 13. OWL-S 1.1 Release, http://www.daml.org/services/owl-s/1.1/ 14. Zhang, Z., Zhou, D., Yang, H., Zhong, S.: A Service Composition Approach Based on Sequence Mining for Migrating E-learning Legacy System to SOA. International Journal of Automation and Computing

A Study of Formative Assessment Index System for Educational Technology Competence Based on AHP Kefei Wang and Lu Ming Jilin Business and Technology College, Jilin Changchun, 130062 [email protected]

Abstract. Formative assessment is a systematic evaluation method that improves the professional standards continuously and gets feedback constantly during the process of education and teaching, so as to improve the teaching. It carries out in the education and teaching activities with the objective is to identify deficiencies in the work of teachers, in order to provide a basis for continuous improvement of teaching. In this paper the author constructs 17 indices including two levels and uses AHP as an indicator to complete each weight distribution and lays a foundation for establishing a scientific, reasonable and feasible evaluation index system. Keywords: Educational technology ability, Formative assessment, Index system, Weight, AHP.

1 Introduction At present, majority of primary and secondary school teachers urgently requires improve the educational technical application ability in the reform of Chinese basic education. Teachers need to have been enquired to use information technology effectively and reform education style, then it can change the way the students learn. In another aspect, educational information also depends on the capacity of teachers to apply educational technology, the higher capacity, the better information technology. National Ministry Of Education issued “Educational Technology Competency Standards for primary and secondary school Teachers” at December 25, 2004,( Hereinafter referred to as "standard")[1]. To promote the implementation of the national primary and secondary teacher in technical skills we will conduct the training, examination, and certification from 2007. We should perfect the system construction progressively by the year 2010. With the training program gradually expanded how to monitor the training quality effectively and assert the quality of what the trainee learnt ? In various forms of training, we find the evaluation of trainee learning effect should not depend on the terminal examination results[2]. We will establish formative assessment index system according to the implementation of the training process, at the same time, considering the impact of various factors that educational technology competence. Weights of the various indices will impact on the evaluation of the effectiveness directly. Therefore, the scientific and reasonable weight is very important that can determine the success or failure of the entire evaluation system. X. Zhang et al. (Eds.): Edutainment 2010, LNCS 6249, pp. 337–344, 2010. © Springer-Verlag Berlin Heidelberg 2010

338

K. Wang and L. Ming

2 Structure and Content of Educational Technology Competence There are the following features in the formative assessment index system: the first, the evaluation process is very important. The process of education is better concerned about than focusing on results. Second, the teachers themselves are not only evaluated, but also they will participate in the evaluation of others, moreover, the later is the primary feature in this system[3]. The third, the conclusions of evaluation also are not only examination of the past work and institutional arrangement of rewards and punishment, but also want to provide diagnostic advice for future work. The last, the formative assessment index system will reflect the spirit of democracy and humanistic better. So, Our technical skills training not only collect the information from the trainees attitude, but also the material that can demonstrate the spirits of collaboration, as well as the contribution for class and team in training process. Eventually according to the overall target of this training we will focus on the instructional design aspects of the program. That is requiring the trainees to master instructional design methods, thinking, learning outcomes and other information collected. At last all these factors are combined and we will construct the formative assessment index system for the primary and secondary teachers referring to authoritative comments and suggestions, which includes the daily behavior, instructional design capabilities, team collaboration and completion results.

3 The Use of Technical Skills Training for Teachers in Formative Evaluation Iindex Weights with AHP in Primary and Secondary Education In the completed and reasonable evaluation of the formation of indicators are an important prerequisite for evaluation and examination, but the real difficulty is to determine the weight of each index. The domestic and foreign scholars have done many studies about how to identify index for assessing the weight. There are two popular method used widely: expert-score method, and another is Analytic Hierarchy Process (AHP). Comparatively speaking, the former method is relatively simple and feasible, but somewhat less accuracy and the latter approach is more scientific, but there are some difficulties in the implementation process for some experts supporting. In this paper, we use the second method --AHP. 3.1 Brief Introduction of AHP Analytic Hierarchy Process (in short: AHP) will break the elements into various levels: objectives, guidelines, programs. These elements are related to decision-making. The qualitative and quantitative analysis method will be adopted on the basis of ones. The method proposed by Pittsburgh University, Satie professor, who is an expert in operational research to use widely in decision analysis of hierarchical weights .At the same time, he is making research for U.S. Department of Defense, which is about “the distribution power to each industrial sector according to their various contribution to national welfare” .He applied the network system theory and multi-objective comprehensive

A Study of Formative Assessment Index System

339

evaluation method and then proposed a hierarchical weights decision analysis method. The method decomposes some complex problems into multiple elements that will be further decomposed according to their relations of domination. Then it will construct a multi-objective, multi-level model by the order target layer, criteria layer and indicators layer. Finally it will get ordered hierarchical level. The method determines the relative importance in all the various factors by neighboring comparative method, and then determines overall order of the relative importance of various factors considering a comprehensive assessment for the subject. AHP separates complicated decision-making system to some levels. It converts multiple elements of complicated problems that is often considered as a whole into neighboring comparison in all related elements, and then continue to convert into the overall weight sort problems. At last we will determine the weights of index. The specific flow chart is as follows:

Comprehensive evaluation index system

Evaluation index hierarchy construction

Construct the matrix by neighboring comparison com-

Levels of single-sorted, calculate weight

Consistency check, CR=CI/RI