Using Learning Style-Based Diagnosis Tool to Enhance Collaborative Learning in an Undergraduate Engineering Curriculum

Using Learning Style-Based

Diagnosis Tool to Enhance

Collaborative Learning

in an Undergraduate

Engineering Curriculum

CHENN-JUNG HUANG,

JIA-JIAN LIAO,

HUNG-YEN SHEN,

NWE NI AYE,

YU-WU WANG,

HONG-XIN CHEN,

DIAN-XIU YANG,

YUN-CHENG LUO,

YI-TA CHUANG

1

Department of Computer Science and Information Engineering, National Dong Hwa University, Hualien, Taiwan

2

Institute of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan

3

Department of Computer Science and Information Engineering, National Tsing Hua University, Hsinchu, Taiwan

4

Department of Computer Science and Information Engineering, National Chiao Tung University, Hsinchu, Taiwan

Received 28 February 2007; accepted 18 March 2009

ABSTRACT: In this study, an intelligent learning style aware diagnosis agent for computer-supported cooperative learning is proposed. Learners are first assigned to heterogeneous groups based on their learning styles questionnaire given right before the beginning of learning activities on the e-learning platform. The proposed diagnosis agent then scrutinizes each learner’s learning portfolio on e-learning platform and automatically issues feedback messages in case some learner’s behavior that is unfitted to his/her learning styles or devious argument on discussion board or wiki is detected. The Moodle, an open-source software e-learning platform, is used to establish the cooperative learning environment for this study. The experimental results reveal that the proposed learning style aware diagnosis agent indeed boosts the performance of the learners.ß 2009 Wiley Periodicals, Inc. Comput Appl Eng Educ 19: 739
746, 2011; View this article online at wileyonlinelibrary.com/ journal/cae; DOI 10.1002/cae.20359

Keywords: e-learning platform; learning style; computer-supported cooperative learning; diagnosis; feedback; weighted decision graph

INTRODUCTION

In the past few years, developing useful learning diagnosis and assessment systems has become a hot research topic in the literature [1
8]. As the Internet gains wide popularity around the world, e-learning is taken by the learners as an important study aid. In order to help teachers easily analyze students’ portfolios in

an intelligent tutoring system, many researchers try to transform students’ portfolios into some useful information, and hopefully reflect the extent of students’ participation in the curriculum activity.

Recently, a lot of e-learning platforms can be used for free on the Internet due to the increasing availability of open-source software. In this work, we incorporate learning parameter improvement mechanisms into an Open Software e-learning platform, Moodle (http://moodle.org/), to verify the feasibility of the proposed algorithms. The most important reason for the choice of Moodle as the pupils’ online learning platform is that Moodle is used for free and is designed to support a social constructionist framework of education, such as collaboration, activities, critical reflection, etc. Meanwhile, there are several

Correspondence to C.-J. Huang (cjhuang@mail.ndhu.edu.tw). Contract grant sponsor: National Science Council of the Republic of China, Taiwan; Contract grant numbers: NSC 96-2628-E-259-022-MY3, NSC 97-2218-E-259-005.

ß 2009 Wiley Periodicals Inc.

striking features in Moodle that well suit the pupils’ online learning in this work:

* _{It is suitable for 100% online classes as well as}

supplementing face-to-face learning.

* _{It supports simple, lightweight, efficient, compatible,}

low-tech browser interface.

* _{Most text entry areas, such as resources, forum postings,}

etc., can be edited using an embedded WYSIWYG HTML editor.

* _{It allows flexible array of course activities—Forums,}

Quizzes, Resources, Choices, Surveys, Assignments, Chats, Workshops.

* _{The teachers can define their own scales to be used for}

grading forums and assignments.

* _{It is effortless to proceed with further study and analysis on}

the learners’ profiles as they are comprehensively estab-lished.

The synchronous chat tool was used and learners in a small assigned group analyze the problem together. Based on their prior knowledge, they determined the information they already had and what information they were still required to possess and had to learn to solve the problem. During this collaboration, they propose hypotheses to the problem; organize a plan of action required to tackle the generated learning issues; and assign group members to conduct defined tasks. Learning takes place in an active and interactive environment, and the learner constructs knowledge by formulating ideas into words with these ideas built upon the reactions and responses of others.

Traditional techniques for detecting similarity between long texts documents have focused on analyzing shared words [9]. Such techniques are usually effective when dealing with long texts because similar long texts usually contain a degree of co-occurring words. However, word co-occurrence in short texts may be scarce or even nonexistent mainly because of the inherent flexibility of natural language, which enables people to express similar meanings using quite different sentences in terms of structure and content. Since such word co-occurrence information in short texts is very limited, this problem poses a difficult computational challenge. To tackle this challenge, the focus of this paper is primarily on developing algorithms to compute the similarity between very short texts, and then apply the algorithms to detect the relevance degree between the messages posted on discussion board and wiki and the learning topics. Meanwhile, some proper feedback messages will be issued to the learners in case some abnormal behaviors such as idleness are detected.

It has been proven that teaching the learners according to their learning styles could effectively assist the learners in the learning activities. There have been several learner’s learning style models proposed in the literature. Kolb/McCarthy learning cycle [10
13] is based on the assumption that learning involves a cycle of four learning stages, but each learner is likely to feel most comfortable in one of the four stages of the cycle based on his/ her preferences along two dimensions of Perception (abstract/ concrete) and Processing (active/reflective). There are five learning style dimensions specified in Felder
Silverman learning style model [14
19], which are Perception (sensing/ intuitive), Processing (active/reflective), Input (visual/verbal), Organization (inductive/deductive), and Understanding (sequen-tial/global). Gee [20] examined the influence of student learning

style preference on student achievement in course content, course completion rates, and attitudes about learning in an on-campus or distance education remote classroom. Both distance and on-campus groups were taught simultaneously by the same instructor, received identical course content, and both groups met weekly. The Grasha
Riechmann Student Learning Style Scales (GRSLSS) has been used to identify the preferences learners have for interacting with peers and the instructor in the classroom setting [21
24]. The six social learning styles identified by GRSLSS are the Independent, Dependent, Compet-itive, Collaborative, Avoidant, and Participant. The inventory is made up of 60 items, including 6 scales, 10 items per scale. Learners are asked to judge themselves using a five-point rating scale that ranges from strongly disagreement to strongly agree-ment. Among the different learning style instruments, the GRSLSS seems to be ideal for assessing student learning preferences in a college-level distance learning setting. First, the GRSLSS is one of the few instruments designed specifically to be used with senior high school and college/university students. Second, the GRSLSS is a relevant scale that addresses one of the key distinguishing features of a distance class, the relative absence of social interaction between instructor/learner and learner/learner. Third, the GRSLSS promotes a desirable learning environment by helping faculty design courses and develop sensitivity to learner needs. Fourth, the GRSLSS promotes understanding of learning styles in a broad context, spanning six categories. Learners possess all of six learning styles, to a greater or lesser extent, and a rationale for pursuing personal growth and development in the underutilized learning style areas is thereby provided.

A brief discussion of each learning style is enumerated below.

* _{Independent students prefer independent study, self-paced}

instruction, and would prefer to work alone on course projects rather than with other students.

* _{Dependent learners look to the teacher and to peers as a}

source of guidance and prefer an authority figure to tell them what to do.

* _{Competitive learners attempt to perform better than their}

peers do and to receive recognition for their academic accomplishments.

* _{Collaborative learners acquire information by sharing and}

by cooperating with teacher and peers. They prefer lectures with small group discussions and group projects.

* _{Avoidant learners are not enthused about attending class}

activities and discussion.

* _{Participant learners are interested in class activities and}

discussion, and are eager to do as much class work as possible. They have a strong desire to meet teacher expectations.

Cooperative learning [25
31] has received increased attention in recent years due to the movement to educate learners with disabilities in the least restrictive environment. It is a relationship in a group of learners that requires positive interdependence, individual accountability, interpersonal skills, face-to-face promotive interaction, and processing. Computer-supported cooperative learning (CSCL) [25
36] has grown out of wider research into computer-supported cooperative work (CSCW) and cooperative learning. The purpose of CSCW is to facilitate group communication and productivity, and the purpose

of CSCL is to scaffold students in learning together effectively. They both are based on the promise that group process and group dynamics are supported and facilitated by computer-supported systems in ways that are not achievable by face to face.

In this work, questionnaire developed by James and Gardner [23] is used to determine students’ learning styles. Meanwhile, learners are grouped to facilitate learning based on Kumar’s framework, which examines the relationship of instructional process and information technology to students’ learning styles [24].

The remainder of this paper is organized as follows. The second section shows the details of the intelligent diagnosis agent. The third section reviews and discusses the experimental results. Conclusions and the future work are made in the fourth section.

ARCHITECTURE OF E-LEARNING PLATFORM

There are three major components in the proposed e-learning platform as shown in Figure 1. They are grouping module, curriculum support module, and learning diagnosis agent. The grouping module is used to group the learners with different learning styles according to the questionnaire of learning

styles categorization designed by Kumar [24]. Each learner can be classified into one of the five learning style clusters as shown in Table 1, and each group consists of five members that are selected from the five corresponding learning style clusters. The advantage of grouping the learners this way is because it is expected, for example, that ‘‘avoidant’’ learners can be motivated and pushed forward by ‘‘participant’’ and ‘‘competitive’’ learners; and ‘‘collaborative’’ learners and ‘‘independent’’ learners can be the sources of guidance for ‘‘dependent’’ learners and give timely assistance to ‘‘dependent’’ learners during the progress of the learning activities.

The curriculum support module assists the instructor and the learners in collecting the supplementary learning material and allows the instructor to organize the discussion issues on the discussion board and wiki. The learning diagnosis agent is used to issue the feedback messages to the learners or the instructor in case the transcript posted by the learner is determined to be irrelevant to the designated discussion issues, or some abnormal behaviors of the learner, such as having been idling for some period of time, are detected. Meanwhile, the diagnosis agent will also request the learners who possess collaborative and participant learning styles to pay attention to the learners who stray from the assigned learning activity, and give assistance to the frustrated learners if necessary.

Learning Diagnosis Agent

The learning diagnosis agent diagnoses learner’s behavior in the e-learning platform, and gives appropriate feedback messages if necessary. The most important function of the learning diagnosis agent is to estimate the relevance degree between the posted transcripts and the learning topics on the discussion board and wiki. As shown in Figure 2, the keyword extractor module is first used to separate and extract the Chinese words. Then a language-independent text analyzer, weighted decision graph, is used to compute the relevance degree between the content of each sample transcript and that of the transcript posted by learner. Idleness detector module is used to report the abnormal behaviors of the learners. Finally, the Feedback Module gathers all available information in the learning portfolio database and gives learners appropriate feedback messages. The above-mentioned modules used in the learning diagnosis agent are elaborated as follows. Keyword Extractor. Since the learner might use a word that has the same meaning as some word used in the sample transcripts saved in the database, we built a synonym database that can assist us in checking the words used by the learner, and replace them with some appropriate synonym in the database accordingly.

Chinese Knowledge and Information Processing (CKIP) system developed by Academia Sinica in Taiwan is used to

separate the Chinese words. Meanwhile, verbosity exclusion module is employed to extract verbs and nouns out of the segmentation results for further processing.

Weighted Decision Graph. Decision graph is an extension of a very well-known decision tree representation. Similar to decision trees, a decision graph contains attribute and decision nodes, where attribute nodes contain some kind of test of attributes’ values and decision nodes serve to predict the solution. However, the decision graph principle is more flexible and more general than a decision tree. Since it also contains cycles, additional internal variables can be added to process input in a time-series manner, and makes the decision graphs especially appropriate to deal with signals and continuous data.

Decision graph is used to calculate the similarity between two alphabetical strings S and S0. Each string is separated into ordered sub-strings, and each sub-string corresponds to a node in decision graph, which can assist in determining whether there is a similar sub-string in the original strings S and S0.

A matrix Mm,n, which is used to describe the decision graph

for strings S and S0, is first constructed as follows:

Mm;n¼ e0;0    e0;n .. . ._{. . ..}. em;0    em;n 0 B @ 1 C A ð1Þ

where m and n denote the length of alphabetical string S and that of alphabetical string S0, respectively. Besides, the elements at the first column and the first row in Mm,nare initialized to zeroes.

The value of the element ei,jin Mm,nis determined by

ei;j¼ ei
1;j
1þ 1 ifsi¼ s

0 j

maxðe_i
1;j; e_i;j
1Þ ifs_i6¼ s0_j

ð2Þ where i ¼ 1, . . ., m, j ¼ 1, . . ., n, siis the ith alphabetical sub-string

in string S, and s0_jis the jth alphabetical sub-string in string S’.

Table 1 The Five Learning Style Clusters

Learning styles

Cluster 1 Dependent, Participant, and Competitive Cluster 2 Participant, Dependent, and Collaborative Cluster 3 Collaborative, Participant, and Independent Cluster 4 Independent, Collaborative, and Participant Cluster 5 Avoidant

In this work, we allow the instructor to grade each sample transcript recorded in the database within the commonly adopted range 0
100. The traditional decision graph should be modified accordingly to accommodate the need for weighting each sample during the process of estimating the relevance degree between the posted transcripts and the expected answers for the discussion issue.We first construct a matrix Wm,nwhich is used to describe

the weighted decision graph for the posted transcript T and the sample transcript T0as follows:

Wm;n¼ e0;0    e0;n .. . ._{. . ..}. em;0    em;n 0 B @ 1 C A ð3Þ

where m and n denote the length of the posted transcript T and that of the sample transcript T0, respectively. Besides, the elements at the first column and the first row in Wm,n are

initialized to zeroes.

The value of the element ei,jin Wm,nis determined by

ei;j¼ ei
1;j
1þ wj log NS

NKj if si¼ s 0 j

maxðei
1;j; ei;j
1Þ if si6¼ s0

ð4Þ where i ¼ 1,. . .,m, j ¼ 1,. . .,n, sirepresents the ith keyword in the

posted transcript, s0jis the jth keyword in the sample transcript, wj

is a weighting factor which is used to reflect the instructor’s grading on the sample transcripts, NS is the total number of the sample transcripts recorded in the database, and NK_jdenotes the counts of the sample transcripts that contains keyword s0j. The

essential functionality of log(NS/NKj) is to filter out the

redundant word in the sample transcripts.

Based on Equations (3) and (4), the relevance degree between the posted transcript s and the sample transcript s0 is derived by

RD_s;s0¼ em;n

logð m
nj j þ 1Þ þ 1 ð5Þ

where the denominator on the right-hand side of the equation is used to reflect the impact of unequal lengths of the posted transcript and the sample transcript.

Then the relevance degree between the posted transcript and the sample transcript that are most relevant to the posted transcript can be expressed by

C ¼ Max

i RDu;si ð6Þ

where u is the posted transcript, and siis the ith sample transcript

in the database.

The following example illustrates how weighted decision graph computes the relevance degree between the posted transcript and the sample transcripts. We assume that the discussion topic designated by the teacher is ‘‘Why does Token-Ring technique outperform Ethernet when the network is severely congested?’’ The sample transcripts recorded in the database and their scores given by the teacher are listed in Table 2. Now assume the transcript posted by the learner is ‘‘The token passing algorithm adopted by Token-Ring can avoid the collision because the computer must first own the token before they can transmit the packets.’’ Table 3 lists the computation result of Equations (4) and (5). Notably, em,n represents the

cumulative weight for the extracted keywords in the posted transcript that match the keywords in sequence in the sample transcripts, and RD_s,s’computes the relevance degree between the posted transcript and the sample transcripts after taking the lengths of the transcripts into consideration. Based on Equation (6), the second sample transcript as given in Table 2 is determined to be most relevant to the posted transcript as expected. Idleness Detector. The idleness detector is used to discover whether the learner has not participated in group discussion for some period of time. Figure 3 shows an example of feedback message that reminds or encourages learners to give their opinions on the discussion board or wiki when the idleness detector finds out the learner has been doing nothing for over 5 min.

Feedback Module. The feedback module may issue messages to the learners/instructor if it is activated by weighted decision graph or idleness detector. A set of feedback rules built in the feedback

Table 2 The Sample Transcripts Recorded in the Database and Their Scores Given by the Teacher

Sample transcript Score given by the teacher In Ethernet, a computer stops transmitting packets when it observes the medium is busy. No computer can send the

packets when the network is congested because the medium is always busy

60 In Token-Ring, a computer can only use the network when it owns the token. This can avoid collisions because only one

machine can use the network at any given time

100 The CDMA/CD algorithm adopted by Ethernet introduces some overhead. The computational overhead becomes heavy

when the network is congested

0 The token passing mechanism performs better than CDMA/CD when the network is congested 50 Table 3 The Relevance Degree Between the Posted Transcript and the Sample Transcripts

The matched keywords in sequence em,n RDs,s0

Computer, stop, transmit, packet 34.9108399549598 19.633222959761 Token-Ring, computer, own, token, avoid, collision 90.3089986991943 61.1385141271519

N/A 0 0

message database corresponds to the outputs of weighted decision graph and idleness detector. The feedback message can be used to give advices to the learners if the transcript posted by the learner is determined to be irrelevant to the designated discussion issues or deviated from the expected answers recorded in the database. The instructor is also able to receive the message regarding the occurrence of biased group discussion. Mean-while, different feedback rules for corresponding learning styles are collected in the database with the assistance of the instructor. The example message as given in Figure 4 applauds the learners of cluster 3, who possess collaborative, participant, and indepen-dent learning styles, for their correct answers to the question, and encourage them to keep actively involved in group discussion.

EXPERIMENTAL RESULTS

To verify the effectiveness of the diagnosis agent proposed in this work, a freshman course offered at Computer Science Depart-ment, National Hualien University of Education participated in the experiments. There are 32 students enrolled in this class. A pre-test in the form of question and answer was taken after classroom teaching, and nine students passed the test. Notably, short essay questions are given in order to assess students’ understanding of and ability to think with subject matter content. Essay questions are chosen over other forms of assessment in this work because essay items challenge students to create a response rather than to simply select a response and thus can reveal students’ abilities to reason, create, analyze, synthesize, and evaluate [36]. The 32 students were divided into heterogeneous discussion groups in the e-learning activity according to their learning styles. After their discussion on the learning topics, each student was asked to take a post-test in the form of question and

answer again. The post-test is similar in content and difficulty level to the pre-test. It can be seen from the test results, as given in Table 4, that the learners indeed benefited by group discussion held in the e-learning activity.

Better student behavior was also observed in the experiment when compared to normal education. For instance, the students were able to develop skills to become active participants in the learning process and to develop collaborative skills with the assistance of the proposed learning diagnosis agent. A change of a learning preference as in the instance of switching from being dependent learners to independent learners was observed. Mean-while, the proportion of the students distracted from group discussion activity substantially lowered down with the aid of the feedback model.

CONCLUSION AND FUTURE WORK

In this work, a learning style aware learning diagnosis agent based on machine learning techniques is proposed. Learners’ transcripts on discussion board and wiki were examined to detect whether the learners plan wrong solutions. A feedback rule construction mechanism is used to issue feedback messages to the learners in case the diagnosis agent detects that the learners go in the biased direction or they have not been joining the group discussion for some period of time. The experimental results exhibit that the proposed work is effective in assisting the students with different learning styles in learning the database concepts taught at a freshman course.

In the future work, we plan to employ advanced machine learning techniques to access each individual’s performance based on the contents of the learners’ transcripts. Meanwhile, we will try to automatize feedback message rule construction mechanism to further reduce the instructor’s labor work. In addition, more advanced text mining technique will be adopted in our future work to accurately measure the relevance degree of expected solutions and posted transcripts.

Figure 3 An example of the message issued by idleness detector.

Figure 4 An example of feedback message for the learners of cluster 3.

Table 4 The Comparison of the Experimental Results of Pre-Test and Post-Test

Pre-test Post-test Pass 28.12% (9/32) 96.87% (31/32) Fail 71.88% (23/32) 3.13% (1/32)

ACKNOWLEDGMENTS

The authors would like to thank the National Science Council of the Republic of China, Taiwan, for financially supporting this research under Contract Nos. NSC 96-2628-E-259-022-MY3 and NSC 97-2218-E-259-005.

REFERENCES

[1] J. Zhang, B. Cheung, and L. Hui, An intelligent tutoring system: SmartTutor, Proceedings of 2001 World Conference on Educational Multimedia, Hypermedia and Telecommunications, June 2001, pp 2130
2131.

[2] B. Cheung, L. Hui, J. Zhang, and S. M. Yiu, SmartTutor: An intelligent tutoring system in web-based adult education, J Syst Softw 68 (2003), 11
25.

[3] S. Y. Cheng, C. S. Lin, H. H. Chen, and J. S. Heh, Learning and diagnosis of individual and class conceptual perspectives: An intelligent systems approach using clustering techniques, Comput Educ 44 (2005), 257
283.

[4] G. Tsaganou, M. Grigoriadou, C. Theodora, and D. Koutra, Evaluating an intelligent diagnosis system of historical text comprehension, Expert Syst Appl 25 (2003), 493
502.

[5] J. J. Lo, H. M. Wang, and S. W. Yeh, Effects of confidence scores and remedial instruction on prepositions learning in adaptive hypermedia, Comput Educ 42 (2004), 45
63.

[6] C. Depradine, Expert system for extracting syntactic information from Java code, Expert Syst Appl 25 (2003), 187
198.

[7] H. C. Huang and T. Y. Wang, A learning diagnosis architecture with a Bayesian network approach, Fifth IEEE International Conference on Advanced Learning Technologies, July 2005, pp 33
34. [8] C. M. Chen and Y. L. Hsieh, Mining learner profile utilizing

association rule for common learning misconception diagnosis, Fifth IEEE International Conference on Advanced Learning Technologies, July 2005, pp 588
592.

[9] G. Salton, Automatic text processing: The transformation, analysis, and retrieval of information by computer, Addison-Wesley Long-man Publishing Co., Inc. Boston, MA, 1989.

[10] W. J. Rong and Y. S. Min, The effects of learning style and flow experience on the effectiveness of e-learning, Fifth IEEE Interna-tional Conference on Advanced Learning Technologies, 2005, pp 802
805.

[11] L. H. Teresa and D. D. Budny, Research on learning style: Applications in the physics and engineering classrooms, IEEE Trans Educ 44 (2001), 276
281.

[12] J. E. Sharp, Teaching teamwork communication with Kolb learning style theory, 31st Annual Frontiers in Education Conference, Vol. 2, 2001.

[13] J. D. Adams, B. S. Rogers, and L. J. Leifer, Microtechnology, nanotechnology, and the scanning-probe microscope: An innovative course, IEEE Trans Educ 47 (2004), 51
56.

[14] P. Paredes and P. Rodriguez, The application of learning styles in both individual and collaborative learning, Sixth International Conference on Advanced Learning Technologies, July 2006. [15] R. M. Felder and L. K. Silverman, Learning and teaching styles in

engineering education, Eng Educ 78 (1988), 674
681.

[16] R. M. Felder, Reaching the second tier: Learning and teaching styles in college science education, J Coll Sci Teach 23 (1993), 286
290. [17] R. M. Felder and E. R. Henriques, Learning and teaching styles in foreign and second language education, Foreign Lang Ann 28 (1995), 21
31.

[18] R. M. Felder and J. E. Spurlin, Applications, reliability, and validity of the index of learning styles, Int J Eng Educ 21 (2005), 103
112. [19] R. M. Felder and R. Brent, Understanding student differences, J Eng

Educ 94 (2005), 57
72.

[20] D. G. Gee, The impact of students’ preferred learning style variables in a distance education course: A case study. Eastern New Mexico University, Portales, 1990.

[21] A. Grasha, Teaching with style. Alliance, Pittsburgh, PA, 1996. [22] S. Hruska-Riechmann and A. F. Grasha, The Grasha-Riechmann

student learning style scales: Research findings and applications. In: J. Keefe (Ed.), Student learning styles and brain behavior, NASSP, Reston, VA, 1982.

[23] W. B. James and D. L. Gardner, Learning styles: Implications for distance learning, New Directions for Adult and Continuing Education 67 (1995), 19
32.

[24] P. Kumar, A. Kumar, and K. Smart, Assessing the impact of instructional methods and information technology on student learning styles, Issues Informing Sci Inf Technol 1 (2004), 533
544.

[25] F. Fischer and H. Mandl, Being there or being where? Video-conferencing and cooperative learning. In H. Van Oostendorp (Ed.), Cognition in a digital world Hillsdale, Lawrence Erlbaum Associates, NJ, 2002, pp 53
73.

[26] C. Gutwin and S. Greenberg, A descriptive framework of workspace awareness for real-time groupware, Comput Supported Coop Work 11 (2002), 411
446.

[27] A. Bucci and J. Pollack, On identifying global optima in cooperative coevolution, Proceedings of the Genetic and Evolutionary Compu-tation Conference, ACM, 2005, pp 539
544.

[28] S. Kapetanakis and D. Kudenko, Reinforcement learning of coordination in cooperative multi-agent systems, Proceedings of the Nineteenth National Conference on Artificial Intelligence, 2002.

[29] L. Panait and S. Luke, Cooperative multi-agent learning: The state of the art. Auton Agents Multi-Agent Syst 11 (2005), 387
434

[30] L. Panait and S. Luke, Time-dependent collaboration schemes for cooperative coevolutionary algorithms, Proceedings of the 2005 AAAI Fall Symposium on Coevolutionary and Coadaptive Systems, 2005.

[31] B. W. Wang, S. Y. Cao, W. M. Huang, L. Weng, Y. Sun, and S. S. C. Busbridge, Structure, Curie temperature, and magnetostriction of alloys, IEEE Trans Magn 42 (2006), 3123
3125.

[32] A. F. Hadwin, C. L. Z. Gress, and J. Page, Toward standards for reporting research: A review of the literature on computer-supported collaborative learning, Sixth International Conference on Advanced Learning Technologies, 2006, pp 1007
1011. [33] H. D. Rodriguez and O. O. Aguirre, Discovery, dissemination and

integration of knowledge in a CSCL system, Seventh Mexican International Conference on Computer Science, 2006, pp 62
69. [34] E. Nesrine, B. B. Narjes, and B. A. Mohamed, Designing

educational systems for use: Case study of Tunisian primary schools, 2nd Information and Communication Technologies, Vol. 1, 2006, pp 66
71.

[35] D. Hernandez-Leo, E.D. Villasclaras-Fernandez, J.I. Asensio-Perez, Y.A. Dimitriadis, M.L. Bote-Lorenzo, and B. Rubia-Avi, Linking collaborative learning practice with IMS LD and service-oriented technologies: An approach based on collaborative learning flow patterns, Sixth International Conference on Advanced Learning Technologies, July 2006, pp 1109
1110.

[36] C. M. Reiner, T. W. Bothell, and R. R. Sudweeks, Preparing effective essay questions: A self-directed workbook for educators. New Forums Press, Stillwater, OK, 2005.

BIOGRAPHIES

Chenn-Jung Huang received the BS degree in electrical engineering from National Taiwan University, Taiwan and the MS degree in computer science from University of Southern California, Los Angeles, in 1984 and 1987. He received the PhD degree in electrical engineer-ing from National Sun Yat-Sen University, Taiwan, in 2000. He is currently a professor in the Department of Computer & Information Science, National Dong Hwa University, Taiwan. His research interests include computer communication networks, data mining, and diagnosis agent for e-learning.

Jia-Jian Liao is pursuing a Master’s degree at the Institute of Learning Technology, National Dong Hwa University, Taiwan. His research interests include computer communication networks, data mining, and applications of machine learning techniques.

Hung-Yen Shen is pursuing a Master’s degree at the Institute of Learning Technology, National Dong Hwa University, Taiwan. His research interests include computer communi-cation networks, data mining, and applicommuni-cations of machine learning techniques.

Nwe Ni Aye is pursuing a Master’s degree at the Institute of Networking and Multimedia Technology, National Dong Hwa University, Taiwan. His research interests include computer communication networks, data mining, and applications of machine learning techniques.

Yu-Wu Wang is pursuing a Master’s degree at the Institute of Learning Technology, National Dong Hwa University, Taiwan. His research interests include computer communication networks, data mining, and applications of machine learning techniques.

Hong-Xin Chen is pursuing the PhD degree at the Institute of Computer Science and Infor-mation Engineering, National Taiwan Univer-sity, Taiwan. His research interests include computer communication networks, data min-ing, and applications of machine learning techniques.

Dian-Xiu Yang is pursuing the PhD degree at the Institute of Computer Science and Infor-mation Engineering, National Taiwan Univer-sity, Taiwan. His research interests include computer communication networks, data min-ing, and applications of machine learning techniques.

Yun-Cheng Luo is pursuing the PhD degree at the Institute of Computer Science and Infor-mation Engineering, National Tsing Hua Uni-versity, Taiwan. His research interests include computer communication networks, data min-ing, and applications of machine learning techniques.

Yi-Ta Chuang is pursuing the PhD degree at the Institute of Computer Science and Infor-mation Engineering, National Chiao Tung University, Taiwan. His research interests include computer communication networks, data mining, and applications of machine learning techniques.