COLLABORATION AND RECOMMENDATION FOR GROUP LEARNING USING IN e-NOTEBOOK SYSTEM

全文

(1)COLLABORATION AND RECOMMENDATION FOR GROUP LEARNING USING IN e-NOTEBOOK SYSTEM. XIN WAN. THE UNIVERSITY OF ELECTRO-COMMUNICATIONS MARCH 2011.

(2) Collaboration and Recommendation for Group Learning Using in e-NOTEBOOK System Xin WAN Submitted to The University of Electro-Communications In partial fulfillment of the requirements for the degree Doctor of Philosophy. Graduate School of Information Systems The University of Electro-Communications. March 2011.

(3) Collaboration and Recommendation for Group Learning Using in e-NOTEBOOK System. Approved by Supervisory Committee: Chair: Member: Member: Member: Member:. Professor Toshio OKAMOTO Professor Akihiko OHSUGA Professor Hideki KOIKE Assoc. Professor Hisashi KOGA Assoc. Professor Maomi UENO.

(4) Copyright by Xin WAN 2011.

(5) e-NOTEBOOK システムにおけるグループ学習のための協調と推薦万. 欣. 概要 e-NOTEBOOK システムは、研究開発活動などにおいて情報共有、共同作業、実験・作業記録等で利用するツールである。本研究では、テーマや課題に対する調査・議論・探求を行うようなグループ学習場面で、学習者を支援するためのe-NOTEBOOKシステムを開発した。本研究の目的はWeb 上で有効なグループ学習を行うための学習環境を構築することである。最初に、探求に基づく協調学習モデルを提案した。(1)学習者はWeb上で相関知識を探求しながら、知識を構築していく、(2)ノートやコメントを書くことにより、知識を共有し、有用な学習資源を提供する、(3)学習者の推薦と有用な学習資源の推薦により、知識獲得を促進する。学習者は知識獲得者である一方、知識提供者でもある。個人とグループ学習プロセスをそれぞれモデリングした。第二は、学習者の学習プロセスや学習活動を基に、社会的相互作用分析に基づく上級学習者の推薦を提案した。まず、学習者の学習活動から、Interaction Analysis Model (IAM)に基づき、相互作用指標（Webページの理解度、リマークの適切度とコメントの合意度）を自然言語処理とデータマイニングを用いて、自動的に抽出する。そして、抽出した相互作用指標を用い、グループ内学習者の学習プロセスに基づき、学習者の知識と理解の分散状況を表す比較関係指標を定義する。最後に、マルコフチェーンモデルで定義した比較関係指標を用い、学習者間の比較関係を算出する。それにより、学習者の推薦を行い、協調活動を促進する。第三は、学習の特徴に配慮し、学習プロセスと社会的相互作用の二つのメカニズムから新しい学習資源の推薦方法を提案した。まず、学習者個人の学習プロセスから、キーワードマップに基づくコンテンツフィルタリングを提案した。キーワードマップの構造は探索空間を制限することができ、足りない情報を推測できる。キーワードマップで表す情報はユーザの嗜好データの足りない部分を推定できる。次は、学習者のレベルや能力などを配慮し、グループ学習者の学習プロセスから、学習者関係に基づく協調フィルタリングを提案した。学習者の相関関係行列を加えて、学習者に信頼できる学習資源を推薦でき、推薦精度を高めている。最後に、提案した二つの方法を組み合わせて、即ち、個人の学習プロセスとグループ学習者の学習プロセス両方を考えて、有用な学習資源を学習者に推薦し、知識の利用と獲得を促進する。実験により提案した学習モデルと推薦方法の有効性と有用性を証明した。協調と推薦の技術を用いることにより、有効なグループ学習ができたと言える。.

(6) Collaboration and Recommendation for Group Learning Using in e-NOTEBOOK System Xin WAN Abstract In this study, we have proposed and developed an e-NOTEBOOK system which provides an inquiry based collaborative learning web environment. Proposed system allows group learners to save the web pages of interest, to express their opinions and ideas by adding notes and comments, to record their activities, and to communicate and collaborate with others. In our system group learning is characterized by: (1) inquiry based knowledge construction; (2) knowledge production through note-taking and comment-taking; (3) knowledge acquisition through collaboration with the recommended advanced learners and reflective use of recommended learning resources. Proposed advanced learners recommendation is based on analysis of social interactions by focusing on learners participating in online knowledge construction and sharing activities. Firstly, the proposed system automatically extracts and analyzes interaction indicators based on learners’ learning activities and learning processes. Secondly, comparison indicators are defined as to describe settings in which knowledge and understanding are socially distributed among group learners. Finally, proposed system evaluates contributions made by each learner (based on defined comparison indicators) and recommends advanced learners to collaborate with. Since the learners may assume a role of an expert or a novice (depending on topics), the proposed system allows for creating flexible and dynamic collaboration among group learners. The proposed recommendation of learning resources takes into account both of the crucial components: the learning process and social interaction. Firstly content filtering based on keyword map is proposed as to the effectiveness of content based filtering. Secondly collaborative filtering based on learner’s implicit relationships is proposed as to provide asymmetric inter personal influence. These approaches allow for more flexibility and control of the recommendation process, and are especially well suited for e-learning domain. Through theoretical and experimental studies we demonstrate the effectiveness and applicability of the proposed approaches and the system that integrates them..

(7) Table of Contents Table of Contents .......................................................................................................... I Table of Figures ......................................................................................................... IV Table of Tables ........................................................................................................... VI List of important notations and terminology ......................................................... VII 1 Introduction .............................................................................................................. 1 1.1 Pedagogical background and motivation ........................................................................ 2 1.2 Research purpose and scope ........................................................................................... 5 1.3 Proposed approach ......................................................................................................... 9 1.4 Contributions of this thesis ........................................................................................... 15 1.5 Outline of this thesis ..................................................................................................... 17. 2 Related Works ........................................................................................................ 20 2.1 Related works on e-NOTEBOOK systems .................................................................... 21 2.2 Recommender system ................................................................................................... 23 2.2.1 Framework of recommender system....................................................................... 23 2.2.2 Recommender system applications ......................................................................... 26 2.2.3 Related works on content filtering based on keyword map ..................................... 30 2.2.4 Related works on collaborative filtering based on learner’s relationship ................ 32 2.3 Related works on advanced learners recommendation ................................................. 37. 3 e-NOTEBOOK System for Group Learning ......................................................... 40 3.1 Learning theories and conceptual framework ............................................................... 41 3.1.1 Constructivism and inquiry based learning ............................................................ 41 3.1.2 The design of inquiry based collaborative learning................................................. 43 3.2 System architecture ...................................................................................................... 47 3.3 The structure of “note-taking” and “comment-taking” ................................................. 49 I.

(8) 3.4 Modeling of learning process and its utilization ............................................................ 51 3.5 Conclusion .................................................................................................................... 54. 4 Recommendation of Advanced Learners Based on Analysis of Social Interactions .................................................................................................................................... 55 4.1 Overview ...................................................................................................................... 56 4.2 Our approach ............................................................................................................... 57 4.2.1 Defining interaction indicators ............................................................................... 59 4.2.2 Estimating comparison indicators .......................................................................... 67 4.2.3 Estimating learners’ comparison ............................................................................ 71 4.3 Experimental evaluation ............................................................................................... 74 4.4 Conclusion .................................................................................................................... 80. 5 Recommendation of Learning Resources .............................................................. 81 5.1 Framework of proposed recommendation approach..................................................... 82 5.2 Content filtering based on keyword map ...................................................................... 86 5.2.1 Overview................................................................................................................ 86 5.2.2 Our approach ........................................................................................................ 87 5.2.3 Summary ............................................................................................................... 94 5.3 Collaborative filtering based on learner’s relationship ................................................. 96 5.3.1 Overview................................................................................................................ 96 5.3.2 Our approach ........................................................................................................ 97 5.3.3 Summary ............................................................................................................. 101 5.4 Combination............................................................................................................... 102 5.5 Technological experimental evaluation ....................................................................... 106 5.5.1 Experimental design............................................................................................. 108 5.5.2 Results ..................................................................................................................111 5.5.3 Discussion ............................................................................................................ 116 5.6 Beyond recommendation: discussion based on educational aspect .............................. 117 5.7 Conclusion .................................................................................................................. 122. 6 Experimental Evaluation of e-NOTEBOOK System .......................................... 123. II.

(9) 6.1 Introduction ............................................................................................................... 124 6.2 Method ....................................................................................................................... 126 6.2.1 Participant and procedure.................................................................................... 126 6.2.2 Analytical data ..................................................................................................... 127 6.3 Results and discussion ................................................................................................ 130 6.3.1 Results ................................................................................................................. 130 6.3.2 Discussion ............................................................................................................ 140 6.4 Conclusion .................................................................................................................. 141. 7 Conclusions and Future Works............................................................................ 142 7.1 Conclusions ................................................................................................................ 143 7.2 Future works .............................................................................................................. 148. Appendix A: Development environment ................................................................ 150 Appendix B: The main functions in proposed e-NOTEBOOK system ................. 151 (a) Login and main menu ................................................................................................. 151 (b) Note taking ................................................................................................................. 152 (c) Making comment ........................................................................................................ 153 (d) Recommendation of learning resources....................................................................... 154 (e) Recommendation of advanced learners ....................................................................... 155 (f) Functions for administrator......................................................................................... 156 (g) Other functions ........................................................................................................... 157. Appendix C: Unsupervised learning algorithm for classification a review ........... 158 Appendix D: A method for analyzing the utterances ............................................. 159 Bibliography ............................................................................................................ 160 Acknowledgments ................................................................................................... 177 Author Biography ................................................................................................... 179 List of Publications Related to the Thesis .............................................................. 180. III.

(10) List of Figures. Fig.1.1 A framework for analyzing group learning based on activity theory........................... 4 Fig.1.2 Research framework ........................................................................................... 8 Fig.1.3 The structure of note taking and comment taking ................................................... 10 Fig.1.4 Basic learning actions in our system...................................................................... 11 Fig.1.5 The structure of this thesis ................................................................................. 18 Fig.2.1 Framework of a recommender system ................................................................... 24 Fig.3.1 Inquiry based collaborative learning in our e-NOTEBOOK system ......................... 44 Fig.3.2 Main system architecture of our system................................................................. 48 Fig.3.3 GUI of note taking and comment taking ................................................................ 50 Fig.3.4 An illustration of how learning process affects recommendation of the learning resources .................................................................................................................. 51 Fig.3.5 Modeling learning process based on Markov Chain................................................ 52 Fig.4.1 Process of learners’ comparison identification........................................................ 58 Fig.4.2 Comments to the remarks ..................................................................................... 62 Fig.4.3 An example of calculating the value of comment ................................................... 66 Fig.4.4 Example of estimating interaction indicators ......................................................... 66 Fig.4.5 Example of estimating comparison indicators ........................................................ 68 Fig.4.6 Example of estimating learners’ comparison .......................................................... 71 Fig.4.7 Sample screenshot for recommendation of advanced learners ................................. 73 Fig.4.8 Learning activities ............................................................................................... 76 Fig.4.9 The predictive accuracy of advanced learners recommendation ............................... 76 Fig.5.1 Framework for recommendation of learning resources ........................................... 83 Fig.5.2 Learning resource and learning process ................................................................. 84 Fig.5.3 Example of generating keyword map based learner profile ..................................... 91 Fig.5.4 Example of keyword map based learner profile ...................................................... 92. IV.

(11) Fig.5.5 Example of content filtering based on keyword map............................................... 94 Fig.5.6 Conceptual flowchart of LR-CF ............................................................................ 97 Fig.5.7 Example of collaborative filtering based on learner’s relationship ......................... 100 Fig.5.8 Sample screenshot for recommendation of learning resources ............................... 105 Fig.5.9 The result of coverage ........................................................................................ 113 Fig.5.10 The result of perfect predication ....................................................................... 113 Fig.5.11 The results of R based on weighting parameter α, β ............................................ 114 Fig.6.1 Outline of the proposed e-NOTEBOOK system ................................................... 125 Fig.6.2 The number of notes of two groups ..................................................................... 131 Fig.6.3 The number of comments of two groups ............................................................. 131 Fig.6.4 The number of new keywords generated everyday ............................................... 132 Fig.6.5 Products (notes & comments) vs. the number of collaborating with advanced learners .............................................................................................................................. 132 Fig.6.6 Products (notes & comments) vs. learning achievement ........................................ 133 Fig.7.1 Proposed main approaches and keywords ......................................................... 144 Fig.7.2 Inquiry based collaborative learning web environment ......................................... 145. V.

(12) List of Tables. Tab.2.1 The main commercial recommender systems ........................................................ 26 Tab.2.2 Recently recommender systems for e-learning ...................................................... 27 Tab.4.1 Interaction indicators .......................................................................................... 59 Tab.4.2 The criterion of calculating value of comment....................................................... 65 Tab.4.3 Spearman's correlation between indicators and predictive accuracy (settings1) ........ 77 Tab.4.4 Spearman's correlation between indicators and predictive accuracy (settings2) ........ 78 Tab.5.1 Keyword map based learner profile generation algorithm ...................................... 90 Tab.5.2 Comparison of typical recommendation methods ................................................ 102 Tab.5.3 Comparison approaches (1) ............................................................................... 105 Tab.5.4 Comparison approaches (2-1) ............................................................................ 106 Tab.5.5 Comparison approaches (2-2) ............................................................................ 106 Tab.5.6 Comparison approaches (3-1) ............................................................................ 107 Tab.5.7 Comparison approaches (3-2) ............................................................................ 107 Tab.5.8 R comparison of different methods (1) ............................................................... 112 Tab.5.9 R comparison of different methods (2-1) ............................................................ 114 Tab.5.10 R comparison of different methods (2-2) .......................................................... 114 Tab.5.11 R comparison of different methods (3-1)........................................................... 115 Tab.5.12 R comparison of different methods (3-2) .......................................................... 115 Tab.6.1 Means, standard deviations and paired t-test of the pre-test and post-test measures of the learners in the evaluation study ........................................................................... 133 Tab.6.2 Educational evaluation results ............................................................................ 139. VI.

(13) List of important notations and terminology Learning resource:. lrb = {wpa , noteab , cabj } Where wp a is the web page a, noteab is the note which relates to web page a, and c abj is the comments which orient to noteab . b is the primary key.. Learning process: Let U = {u , u ,..., u } denotes the set of all learners and LR = {lr , lr ,..., lr } denotes the set of all 1. 2. m. 1. 2. o. learning resources (items). A feature function, f (LR ) , represents learner u ’s preference for the ui. i. learning resources in LR . A learning process is a sequence of tuples lp =< u , lr > . Learning step ui. i. k. is the index of tuple in a learning process. Each learner ui may have a learning process LP to ui. express her/his preferences. Let lp denotes the j th learning step of learner ui and in this j. ui. learning step the learner u learned learning resource lr . Then, the learning process of learner i. k. ui can be represented as LPui = {lp uij }nj =1 .. Keyword map: Keyword _ Map =< K , R >. which basically consists of keywords. ki ,. and relations. rj .. Where K is a set of keywords and R is a set of relations between the keywords.. ki Î K. and. rj Î R .. Learner profile:. Learneri = ( LPi , KeywordMap i , Arelationshipi , Aratingi ) Where, LPi is the learning process of learner i. KeywordMap i =< K i , Ri > is the keyword map of learner i based on learning process. Arelationship = [ri ,1 ,...,r i ,n ] is the relationship of learner i i. and other learners. Arating = [r1,i ,..., rm ,i ] is the rating which was calculated based on the learning i. process of learner i. n is the number of learner. m is the number of learning resource.. VII.

(14) VIII.

(15) 1 Introduction. 1 Introduction .............................................................................................................. 1 1.1 Pedagogical background and motivation ........................................................................ 2 1.2 Research purpose and scope ........................................................................................... 5 1.3 Proposed approach ......................................................................................................... 9 1.4 Contributions of this thesis ........................................................................................... 15 1.5 Outline of this thesis ..................................................................................................... 17. 1.

(16) 2. Chapter 1: Introduction. Computer supported collaborative learning (CSCL) is an emerging research field which focuses on how collaborative learning, supported by technology, can enhance peer interaction and work in groups, and how collaboration and technology facilitate sharing and distributing knowledge and expertise among community members [Lipponen et al., 2004]. The main goal of CSCL is to maximize individual learning based on group work. Many research works on CSCL have tried to discover the factors and design interface that promotes interaction in collaborative and computer-based settings [Kishi and Kubota, 2009]. Extensive research has been done on CSCL; one of the main benefits of CSCL is that it has the potential to generate useful tools for enhancing students' learning and knowledge building [Scardamalia and Bereiter, 1993]. Recently, many CSCL tools such as e-NOTEBOOK systems [Edelson and O’Neill, 1994] [Hewitt and Webb, 1992] [Bell, 1997] [Masukawa, 2004] have been developed and embedded in the e-learning environments.. This dissertation focuses on enabling collaboration through recommendation in e-NOTEBOOK system for group learning. The proposed e-NOTEBOOK system is a WEB based, multimedia database. It is structured to support learners through the inquiry processes and to provide them with a mechanism for working cooperatively with others on the web. Contributions of this dissertation are: 1.. Developed an e-NOTEBOOK system that provides an inquiry based collaborative learning web environment (Chapter 3). System encourages collaboration through recommendation: advanced learners and learning resources.. 2.. Proposed advanced learners recommendation is based on analysis of social interactions of learners participating in online knowledge construction and sharing (Chapter 4).. 3.. Proposed recommendation of learning resources that takes into account both of the crucial components: the learning process and social interaction (Chapter 5).. 1.1 Pedagogical background and motivation. Constructivism is a theory of learning that describes how individuals’ minds create knowledge or how individual knowledge structures their deep conceptual understanding. Recently constructivism.

(17) Chapter 1: Introduction. 3. has become an important paradigm for guiding research and practice in education. There have been an increase in number of e-learning systems that use methods based on constructivism such as inquiry based learning, problem based learning, reciprocal learning etc. Interest in these theories owes much to the philosophical changes such as Piaget’s constructivism, Vygotsky’s social constructivism (e.g. zone of proximal development) and etc. Inspired by these theories, the field of education has recently begun to move from the use of traditional pedagogical approaches to approaches that encourage collaborative, integrated, community-based, elective and learner-centered learning.. In this research, following the principles of constructivism we propose an e-NOTEBOOK system that provides an inquiry based collaborative learning web environment in which group learners learn to conduct collaborative, open-ended investigation on the web. Proposed system allows group learners to save the web pages of interest, express their opinions and ideas by adding notes and comments, record their activities, and communicate and collaborate with others. Inquiry learning is a form of active learning, where progress is assessed by how well students develop experimental and analytical skills rather than how much knowledge they possess. It gives students an opportunity for detailed investigations on worthy topic and enables them to learn from the experiences of others and apply gained knowledge, skills and attitudes to real case in their lives. Our inquiry based collaborative learning emphasizes cooperative learning and learners’ own notes/comments construction to represent what is being learned. Group learners share goals and negotiate the process of creating the notes and comments that helps in increasing both interest and knowledge of learners. Within this framework learners collaborate and work together to make sense of what is happening. In our group learning, learners can define problems, discuss views or predictions, collect information, evaluate collected information, make conclusions, combine views and create a product (note or comment). These tasks involve the learners’ problem solving, decision making and investigative skills. At the same time, our system encourages collaboration through recommendation, such as recommendation of advanced learners and recommendation of learning resources.. The methodological approach taken in the research follows the theoretical framework of activity.

(18) 4. Chapter 1: Introduction. Tool (e-NOTEBOOK). Production Subject/s: Group learners. Knowledge production. Object: Notes & Comments. Outcome. Consumption Knowledge acquisition Exchange Distribution. Rules (inquiry based collaborative learning). Division of Labour Community (group of advanced (not explicitly used) learners during the work placement). Fig.1.1 A framework for analyzing group learning based on activity theory Proposed inquiry based collaborative learning emphasizes cooperative learning and learners’ notes/comments construction to represent what is being learned. In this study, the e-NOTEBOOK system is seen as the activity system in which the group learners are subjects, the notes & comments are the objects and the advanced learner group is the community. Tools are represented by the functions such as note-taking, reading, recommendation etc. provided by proposed e-NOTEBOOK system. The tools mediate the relation between the subjects (group learners) and the objects (notes& comments), the rules (in this study: inquiry based collaborative learning) mediate the relation between the subjects and the community and the division of labor (not explicitly used in this study) mediates the relation between the community and the object. Finally, an activity is motivated by the need to transform the object into an outcome.. theory. Activity theory [Engestrom, 1987] is a powerful framework for the design and development of technology-based learning environments because its assumptions are consistent with the ideas of constructivism, situated learning, distributed cognition and everyday cognition [Jonassen and Rohrer-Murphy, 1999]. Engestrom’s activity theory framework considers how people (subjects) interact with objects via mediating tools in order to affect outcomes. This is done in the context of a community with its own set of rules and approaches for the division of labor. Rules operating in any context or community refer to the explicit regulations, policies, and conventions that constrain activity as well as the implicit social norms, standards, and relationships among members of the community [Jonassen, 2002]. The community consists of the individuals and subgroups that focus at least some of their efforts on the common objective [Engestrom, 1999]. Division of labor refers to both the horizontal division of tasks between cooperating members of the community and the vertical division of power and status [Engestrom, 1999]. In this study, we adopt Engestrom’s framework. As shown in Fig.1.1, this study focuses on how the components of the production (knowledge production) sub-system and the components of the consumption (knowledge.

(19) Chapter 1: Introduction. 5. acquisition) sub-system lead to the accomplishment of the activity-based outcomes. That is to say, this study focuses on tool and community components. The proposed system thus has the potential to constrain and facilitate knowledge production and knowledge acquisition with the tools and functions.. 1.2 Research purpose and scope. In this section, we describe our research purpose and scope, and present research questions.. (1) Research purpose We propose an e-NOTEBOOK system that provides a scaffold for learners in inquiry based collaborative learning on the web. Collaboration in our system is characterized by (1) knowledge construction through inquiry based strategy; (2) knowledge production via the functions of note-taking and comment-taking; (3) facilitation of knowledge acquisition through recommendation. The emphasis of the research is on the learner as an active knowledge producer and the e-NOTEBOOK system as a facilitator and supporter for inquiry based collaborative learning. It provides a scaffold for group learners to engage in collaborative learning. The e-NOTEBOOK system is focused notes and comments. Vygotsky (1978) considered that all artifacts are culturally, historically and institutionally situated. In Vygotsky’s view, interactions with the social environment, including peer interaction and/ or scaffolding are important ways to facilitate individual cognitive growth and knowledge acquisition. Through shared workspaces, e-NOTEBOOK system provides the appropriate support to develop collaboration. According to zone of proximal development [Vygotsky, 1978], more knowledgeable people such as the teachers or peers can support learning of less knowledgeable ones. Therefore, one of the purposes of our research is modeling collaboration and analyzing interaction for understanding learning process and identifying advanced learners to collaborate with.. In general, merely providing learning resources to learners is not enough. Attention needs to be paid to the design of the learning tasks and learning environment so that active use of learning resources.

(20) 6. Chapter 1: Introduction. can be promoted. Since the presence of learning resources does not automatically improve learning, it is necessary to achieve the reflective use of these materials. On the other hand, to help learners to find out each other’s relevant contributions is advantageous for making group learning more effectively as the number of learners increases. Moreover, in group learning, learners are different, have different tastes and preferences, and learn in different ways. At the same time, since the learning resources (products) provided by different learners are increasing exponentially, it becomes difficult for learners to search for useful learning resources (products). [Kirschner et al., 2006] have suggested that student-centered instructional strategies such as discovery learning, problem-based solving, experiential learning, and inquiry-based strategies do not work effectively unless additional guidance is provided to learners. It is necessary to guide learners in their learning thereby improving reflective use of learning resources for facilitation of knowledge acquisition. This leads to the development of systems that identify individual needs of learners and provide them with specific information that satisfy personalized learning goals. Utilizing recommender systems in the e-learning environment can help in providing an automatic process to support learners in finding suitable materials instead of relying on classmates, tutors and other sources [Resnick and Varian, 1997]. Therefore, making use of the recommender systems in the e-NOTEBOOK system can improve reflective use of learning resources (notes and comments) that can utilized as efficient and effective references for learners. That is to say, recommendation of learning resources can be used for navigating and representing knowledge during learning process, thereby improving reflective use of learning resources.. This dissertation focuses on the above research purpose. Since making use of collaboration and recommendation in e-learning to facilitate knowledge production and knowledge acquisition has become a major issue, our objective is to provide collaboration and recommendation functions in e-NOTEBOOK system for group learning in online web-based learning settings. In the following section, we explain our motivation for focusing on the problems of collaboration and recommendation.. (2) Research scope.

(21) Chapter 1: Introduction. 7. Collaboration is a set of instructions prescribing how learners should interact and collaborate and how they should solve the problem [Dillenbourg, 2002]. During collaboration, learners interact employing self-critiquing (reflection), inquiring and arguing skills; these skills promote the knowledge construction. There remains a strong need for having notes and comments features within e-learning tools that support interactions in group learning settings. For example, to make learners’ activities meaningful and significant it is necessary to involve them in participatory activities. In addition, according to [Lea et al., 2002], designs of collaborative work environments that encourage clear and strong definitions of (a) the group itself and (b) the self-identify of members in relation to the group, have potential to raise productivity of learners. In other words, the system needs functionality for identifying the relationships between all of the learners and recommending advanced learners to collaborate with, thereby achieving learning outcomes of learners.. On the other hand, with the increase of learning resources in e-learning, recommender systems have become an important component of personalized e-learning services and are essential for e-learning providers to remain competitive [Thyagharajan and Nayak, 2007] [Wang et al., 2007]. E-learning recommender systems have considered such components as knowledge, interests, goals, tasks, background, individual traits, context of work [Brusilovsky and Mill’an, 2007]. For these purposes, most recommender systems have applied or adopted existing recommendation algorithms, such as collaborative filtering (CF), content-based filtering (CBF). However, recommendation of learning resources is somewhat different from other domains (e.g. e-commerce). The proposed recommendation approach is hybrid and it takes into account both of the crucial components: learning process and social interaction. In addition, instead of asking learners to explicitly rate the learning resources, we track learners’ preferences through implicit means by monitoring their learning processes and learning activities such as note taking, and note reading. Providing e-learning recommendation is a complex processes. The processes includes estimating interests and background of not only the learner himself but also of the advanced learners, and recommending more suitable learning resources to learners as to help them to improve reflective use of learning resources. In e-learning domain, recommender system needs to improve the “educational provision” [Drachsler et al., 2008] [Drachsler, 2009]. Current recommendation techniques for e-learning suffer from many.

(22) 8. Chapter 1: Introduction. e-NOTEBOOK System for Group Learning. Learning Process. Group. Comparison of Learners Estimation. Learners’ comparison matrix. Collaborative filtering based on learner’s relationship. Recommendation based on group. Recommendation of advanced learners. Hybrid Prediction Component. Individual Content filtering based on keyword map. Recommendation of learning resources. Recommendation based on individual. Fig.1.2 Research framework This research consists of three parts: first (yellow layer), represents developed e-NOTEBOOK system to provide the inquiry based collaborative learning web environment. Second (green layer), represents proposed new approach to recommending advanced learners based on learning process and analysis of social interactions by focusing on learners participating in online knowledge construction and sharing. Thirdly (blue layer), proposed recommendation of learning resources takes into account both of the crucial components: the learning process and social interaction. Rectangles represent input or output information; trimmed rectangles represent our proposed system and system processes.. profound problems due to: incomplete user profiles, inadequate mechanisms for representing the information or knowledge of the items and disregarding learning process and social interactions of learners.. We tackled the problem of designing and developing an online web-based e-NOTEBOOK system for group learning based on activity theory. In order to facilitate knowledge production, the system recommends suitable learning resources to learners for reflective use based on their preferences. Meanwhile, the system is capable of identifying the knowledge level of each learner through and analysis of collaboration and social interactions, and at the same time, recommending advanced learners to collaborate with. The system provides a scaffold for learners to collaborate with each.

(23) Chapter 1: Introduction. 9. other and in particular to collaborate with advanced learners. This can facilitate knowledge acquisition that leads to improving learning outcome. In summary, the primary questions we address in this thesis are: 1. How to provide an inquiry based collaborative learning web environment for group learning? 2. How to estimate useful social interaction indicators automatically? How to estimate learners’ comparisons for recommending advanced learner to collaborate with? 3. How to provide recommendation of learning resources? 3-1. How to construct more complete user profiles, and more adequate mechanisms for representing the information of the items (learning resources, notes)? How to use this information to help learners complete their learning goals? 3-2. How to recommend learning resources with regards to learners’ understandings and knowledge levels based on social interactions with asymmetric relationships? How to track the learners’ preference with implicit methods?. 1.3 Proposed approach. In this thesis, we propose an inquiry based collaborative learning web environment that considers the problems mentioned above to provide tools for collaboration and recommendation (Fig.1.2). The system is designed based on the principles of the constructivist learning theories and is focused on providing an effective and efficient scaffold for group learning.. We have developed an e-NOTEBOOK system that provides an inquiry based collaborative learning web environment. Implemented system for group learning aims to address the first question-collaboration. Let us briefly describe the proposed web based e-NOTEBOOK system..

(24) 10. Chapter 1: Introduction. According to Piaget’s definitions [Piaget, 1954, p.19, 64, 110], “knowledge is an interaction between subject and. object…”. “Knowledge…is a. perpetual construction made by exchanges. between…thought and its object…” “Knowledge… isn’t a copy of reality… it’s a reconstitution of reality by the concepts of the subject, who, progressively and with all kinds of experimental probes, approaches the object without ever attaining it in itself.” Knowledge is not independent of the knower; knowledge consists of physical and abstract objects in our experience. These statements imply that the construction of knowledge is a dynamic process that requires the active engagement of the learner. Fig.1.3 is the structure of note-taking and comment-taking used in our e-NOTEBOOK system. Abstract is defined as a sketchy summary of the main ideas or points of the related web page. Remark is defined as something that learners write which expresses opinion, idea, thought, and etc. about the web page. Comment is defined as something that learners write as an explanation, illustration, opinion or criticism of a remark. A Note consists of an abstract and a remark about a certain web page.. WWW Web Page NOTE Abstract. NOTE Abstract. Remark. Remark. Web Page NOTE Abstract ……. Remark. ……. Comments. Fig.1.3 The structure of note taking and comment taking Note-taking is a strategy for making information meaningful; by collecting and recording notes, this information can further be utilized in the transfer of knowledge from one learner’s mind to another. It is a threaded tool which allows learner to post notes (abstract&remark) after searching and finding useful web pages. Our system has a unique implantation of threaded comments which allows learners to post comments to notes and respond to comments already posted, whereby participants create comments and explicitly reply to each other using a “reply to” button. As such, comments are represented by hierarchical, tree structure with a clear flow of messages from the initial parent post to subsequent ideas and thoughts..

(25) Chapter 1: Introduction. 11. Our system is designed to provide a scaffold for group learners as they learn to conduct collaborative, open-ended investigations in collaborative learning. In our system, learners are assigned to groups and provided with an ill-defined and ill-structured problem. Group learners must organize themselves, define objectives, assign responsibilities, conduct research, analyze results, and preset conclusions. The problems are purposely “ill-defined” and “ill-structured”, causing group members to work collaboratively to define specific issues, problems, and objectives. There are five basic learning actions such as ask, investigate, create, discuss, and reflect in our system (Fig.1.4). The inquiry process includes asking questions, investigating solutions, creating, discussing their discoveries and experiences, and reflecting on their new-found knowledge, and asking new questions [Bruce and Bishop, 2002]. For example, suppose that a group of learners want to solve the problem of “Lake Pollution” together by using our system on the web. In order to solve this problem, learners need to search and study a variety of special subjects such as “Nature”, “Chemistry”, “Plant”, “Animal” on the web. This process is referred to as investigating. Since the information on the Internet is neither well-organized nor quality controlled, and web pages are not constructed with. Sharing space. Group. Learning Resources ask investigate. Web. Searching on web. Note taking Recommendation of Advanced learners. create. Web page NOTE Abstract. Recommendation of learning resources reflect. Remark. discuss Comment taking. comment. Fig.1.4 Basic learning actions in our system Given a problem, a learner searches internet (investigate) to find a useful learning resource (ask), and record his opinion about the content (links) of the web pages by writing abstract of the web pages and expressing ideas by creating remarks-notes (create) with the help of the system. In addition, after a learner reads notes and comments of others, he can express his ideas and opinions through comment (discuss/reflect) which in turn allows to construct knowledge base within our system. Our system enables collaboration through recommendation: i.e. learners are recommended learning resources to study and advanced learners to collaborate with..

(26) 12. Chapter 1: Introduction. learning in mind, learners need to process, reconstruct and represent. The learners would need to participate in inquiry based collaborative learning process to construct knowledge through collaboration and constructivist activities such as note-taking, comment-taking and note/comment reading. As learners encounter useful learning resources, they can record the contents (links) of the web pages, write abstracts of the web pages or express their ideas - remarks as notes in our system. When transforming knowledge, it is critical that learners make inference and elaborate on new information by adding details, and generate relationships, both among the new ideas and also between new materials and the information already in memory [Brown et al., 1983]. By trying to explain their ideas to others students and by interacting with their peers on academic content, learners would sharpen their thinking and gain new knowledge. On the other hand, after the learners read the notes or comments of others, they can post or reply their ideas or opinions through comments, and as a result contribute to the construction of knowledge in our system. The learners construct their own knowledge by explaining ideas and approaches based on their prior knowledge and experiences, applying these to a new situation, and integrating the newly gained knowledge. These activities are constructive activities. Learners engage in social self-enhancing behaviors, such as making mutually supportive coordinated contributions (e.g. taking notes). On the other hand, learners also compensate for other group learners’ shortcomings by helping out or doing extra work by writing comments. Notes and comments are contributions and products provided by learners. These products helped learners construct a rich understanding of the problem and provided ideas for solving the problem. The learning activities or participant structure includes learners’ generating their own research questions and knowledge. Learners produce knowledge by taking notes/comments and acquiring knowledge by using notes/comments of others.. Recommendation of advanced learners: Learners can perform group learning in several ways such as (1) Learning by teaching; (2) Learning by diagnosing; (3) Learning by open discussion. Learners may want to find other learners who can help to answer their questions and to obtain the information needed for completing their learning objective. Based on the learners’ learning processes and learning activities we automatically extract interaction indicators by using natural language processing technologies and data mining methods. These interaction indicators are as follows: (1).

(27) Chapter 1: Introduction. 13. Comprehension of Web page (CoW); (2) Adequacy of remark (AoR); (3) Agreement of comment (AoC). We then use these interaction indicators to generate comparison indicators such as indicator of CoW, indicator of AoR, indicator of AoC to estimate comparison of learner’s knowledge levels. These indicators apply to settings in which knowledge and understanding are socially distributed among group learners. We then apply machine learning methods to estimate learners’ comparison matrix with regards to the above mentioned indicators. The focus is on automating the acquisition of group learners’ comparison and analysis of social interactions based on learners’ implicit influence. Finally, the system facilitates knowledge acquisition by automatically recommending collaboration with more knowledgeable (i.e. advanced) learners.. Recommendation of learning resources: The recommendation of learning resources reported in this dissertation addresses the following principal question which is formally defined as follows:. Definition 1.1 : Given an adequate context of the items (here, learning resources such as web pages, notes) I and complete user profiles (including user preference, learner relationship and etc.) which are inferred from users’ behaviors (e.g. learning processes, learning activities such as note taking and reading, comment taking), recommend suitable N items to a user based on her/his profiles, where N Ì I .. This problem is difficult for several reasons. First, users usually have only limited knowledge of the content and structure of the items available in I . This limits how a recommender system can map the user profile. Second, the collection of items I can contains millions of items, but users expect recommendations in less than a second.. Traditional recommender systems have achieved success in many domains; however they are not suitable to the domain of e-learning due to a number of peculiarities. In order to improve the “educational provision”, learner’s educational characters such as learning activities (note taking, note reading, comment taking), learning processes, social interactions need to be considered. By taking these characteristics into account, our recommender system is specifically tailored to.

(28) 14. Chapter 1: Introduction. satisfying educational objectives. By using proposed approaches to realize personalized recommendation based on not only learning histories but also learning activities, we can answer the third question mentioned above.. Let us discuss the proposed approach for learning resources recommendation. Proposed hybrid recommendation approach uses content-based filtering powered by keyword maps, named Content Filtering based on Keyword Map (KM-CBF). We focused our efforts on automatically extracting keyword maps from the text of learning resources provided by learners, which provides a mechanism for representing the information of text. Each keyword in a keyword map is used to represent a subject or object of a sentence. The keywords can be linked through relations and have occurrences linked to them. A keyword map can thus be referred to as a collection of keywords and relations. The fundamental idea here is to use a keyword map to represent textual information. We hypothesize that keyword maps help to increase both the relevance and the complement of learning resources recommendation. Keyword maps provide a form of description of contents of each learning resources (e.g. web pages, notes) and each learner’s existing knowledge with keywords and various relations between them. This structure allows restricting the search space and inferring missing information. The approach described here aims to address the question 3-1 to help learners to complement their learning. We demonstrate the use of keyword maps in a content-based approach to complete the learners’ learning objectives. First, we propose a keyword map based approach, which considers dynamic factors such as learning step of learning process and “transfer of learning” that refers to the expansion and generalization of learning outcomes, to measure the relative importance in learner profile. Second, the recommender module uses the learner profile and the keyword maps of learning resources to create the list of learning resources recommendation based on Jaccard similarity. In our group learning, this mechanism provides inquiry reasoning such as to create stories, analogy making based on individuals’ learning processes.. In addition, we have also proposed a collaborative filtering method based on Learner’s Relationship (LR-CF). We use learners’ learning processes and social interactions to infer the learners’ knowledge levels in order to provide learners’ backgrounds. The result is utilized in the collaborative.

(29) Chapter 1: Introduction. 15. recommendation phase to provide implicit asymmetric inter-personal influence. Firstly, recommendation approach relies upon implicitly acquired behavioral data (e.g. learning step, learning activity) denoting learners’ interests and factors. The factors characterize the learners themselves through learners’ comparison. Secondly we use the multidimensional collaborative filtering based on learners’ implicit influence to provide recommendation of learning resources to every learner in a group. Here we focus on learners’ knowledge level comparisons. Advanced learners have a high probability of giving accurate recommendations to other less advanced learners. In our group learning settings, this mechanism provides collaborative decision making for reflective use of learning resources based on collaborative learning processes. We further combine KM-CBF with LR-CF approach together through CombSUM to perform recommendation. This way, learners can use this system to identify new learning resources based not only on their interests and backgrounds but also on the strengths of others that assist them in learning more effectively and efficiently.. 1.4 Contributions of this thesis. E-learning recommender systems must consider a variety of aspects. Many have already been addressed, such as knowledge, interests, goals, tasks, background, individual traits, and context of work. However, the learning process and knowledge level of the learner are also very important, but have received little attention. In this research, we consider the learning process and knowledge level. The learning process can be seen as accessing learning resources in a particular sequence. We use this definition of the learning process to propose a new recommendation method.. This thesis makes the following contributions:. 1. Proposed and developed e-NOTEBOOK system that provides a scaffold for learners to sharpen their thinking and gain new knowledge in inquiry based collaborative learning. The system facilitates studying and communicating for group learners. It allows learners to: •. Utilize an inquiry based collaborative learning web environment that encourages.

(30) 16. Chapter 1: Introduction. collaboration through recommendation of advanced learners and learning resources. •. Construct knowledge, share knowledge and facilitate knowledge acquisition.. •. Model learning process based on Markov Chain.. 2. Recommendation of advanced learners based on analysis of social interactions. •. Interaction indicators such as (1) Comprehension of Web page (2) Adequacy of remark (3) Agreement of comment are dynamically and automatically extracted and analyzed.. •. The comparison indicators are defined based on interaction indicators, which are symbols that describe a situation in which knowledge and understanding are socially distributed among group learners.. •. Proposed methods for estimating knowledge level comparison relationships between learners, and recommending advanced learners to collaborate with as to promote collaboration.. 3. Proposed recommendation of learning resources takes into account both of the crucial components: learning process and social interaction. •. We have proposed a content filtering based on keyword map, which utilizes individual learning process as to improve the effectiveness of content-based recommendation. It provides essential mechanism that enhances inquiry reasoning in the group learning settings. ². Proposed automatic keyword map generation, which allows to structure information about learning resources and learner preferences.. In addition,. keyword maps allow to restrict the search space and to infer missing information. ². Used keyword map as the foundation for constructing learner profiles and learning resources, to improve recommendation process (even when learner’s explicit preferences are incomplete).. •. Collaborative filtering based on learner’s relationship is proposed based on implicit asymmetric inter personal influence, and the group learning process. Which in turn provides a mechanism for collaborative decision making. ². The system estimates learners’ knowledge level comparison based on defined interaction. indicators,. asymmetric. interpersonal. influence,. and. learner’s.

(31) Chapter 1: Introduction. 17. relationships. It utilizes, learners’ comparison matrix, which represents relations between the learners. Each element of the learners’ comparison matrix describes the relationship between two learners, which is calculated based on learning process and learning activities. ². Proposed to infer learner’s preferences based on step impact coefficient.. ². Proposed multi-dimensional learners’ comparison based on the following dimensions: [User (learner) × Item (learning resource) × User (relationship)] .. These contributions allow us to achieve the objective of providing: (1) Collaboration: an inquiry based collaborative learning web environment, (2) Automatic analysis of interaction used to better understand learning process and to improve collaboration, and (3) Recommendation that is used as a tool of navigating and representing knowledge for effective reflective use of learning resources.. 1.5 Outline of this thesis. Chapter 2 presents the related works. We start with a review of e-NOTEBOOK systems for collaborative learning and outline their advantages and disadvantages. Next, we discuss the related works on learning resources recommendation and discuss their limitations. We review recommender systems and analyze methods employed to make recommendations, and highlight limitations of current recommender systems in the e-learning domain. Subsequently, we describe related works on advanced learners recommendation and discuss their limitations.. Part I Collaboration Chapter 3 introduces proposed e-NOTEBOOK system. Firstly, we present learning theories and proposed learning model of our research. Then, the main architecture of system and main functions are presented. Subsequently, we introduce proposed note-taking and comment-taking modules. Finally, we present modeling of learning process and its utilizations.. Part II Recommendation: It consists of Chapter 4 and Chapter 5..

(32) 18. Chapter 1: Introduction. Chapter 4 presents proposed methods for recommendation of advanced learners based on analysis of social interactions. Firstly, we describe how to dynamically estimate the interaction indicators based on learners’ learning activities. Secondly, we introduce how to estimate comparison indicators based on the interaction indicators. In addition, we describe the meanings of the proposed comparison indicators from the educational viewpoint. In addition, we propose estimating learners’ comparison based on Markov Chain Model, which is utilized for improving the accuracy of recommendation of learning resources. In addition, we demonstrate effectiveness of the proposed approach in different learning settings.. In Chapter 5, we propose recommendation of learning resources, which takes into account both of the crucial components: learning process and social interaction. In section 5.1, the framework of proposed recommendation approach is presented. In the following sections, we describe details of proposed approaches. First, we present the content filtering based on keyword map and describe its use for recommending learning resources based on learner’s individual learning process (section 5.2).. 1. Introduction 2. Related Works. e-NOTEBOOK System 3. Collaboration. Group Learning Process. Individual Learning Process. Recommendation. 4. Recommendation of Advanced Learners. 5. Recommendation of Learning Resources. Content filtering based on keyword map. 6. Experimental Evaluation of e-NOTEBOOK System 7. Conclusion. Fig.1.5. The structure of this thesis. Collaborative filtering based on learner’s relationship.

(33) Chapter 1: Introduction. 19. Second, we introduce the collaborative filtering based on learner’s relationship, and show that it more suitable for implementing learning resources recommendation that takes into consideration not only learning activities and group learners’ learning processes, but also social interactions (focusing on learners’ knowledge level comparison) (section 5.3). Next, a mechanism of combining above mentioned methods is presented (section 5.4). In section 5.5, we present experimental evaluation. In addition, we analyze our approaches from educational aspects and elaborate on impact of proposed approaches (section 5.6).. In Chapter 6, we evaluate the performance of our proposed methods, and the performance of the total system that integrates proposed methods. We then present our views on how to support the learners in group learning effectively and efficiently based on proposed mechanisms. We demonstrate efficiency and impact of proposed approaches from educational aspects. Finally, Chapter 7 presents the conclusions of the thesis and outlines potential future directions..

(34) 2 Related Works. 2 Related Works ........................................................................................................ 20 2.1 Related works on e-NOTEBOOK systems .................................................................... 21 2.2 Recommender system ................................................................................................... 23 2.2.1 Framework of recommender system ...................................................................... 23 2.2.2 Recommender system applications ......................................................................... 26 2.2.3 Related works on content filtering based on keyword map ..................................... 30 2.2.4 Related works on collaborative filtering based on learner’s relationship .................. 32 2.3 Related works on advanced learners recommendation ................................................. 37. 20.

(35) Chapter 2: Related Works. 21. In this chapter we introduce existing e-NOTEBOOK systems (2.1), recommender systems (2.2) and expert finding systems (2.3). In section 2.1, we discuss e-NOTEBOOK systems for collaborative learning and outline their advantages and disadvantages. In section 2.2, we discuss the related works on learning resources recommendation and discuss their limitations such as incomplete user profile, and lack of user reputation, etc. In section 2.3, we describe related works on advanced learners recommendation and discuss their limitations.. 2.1 Related works on e-NOTEBOOK systems Computer supported collaborative learning (CSCL) aims at developing useful tools as to enhance students’ learning and knowledge building [Scardamalia and Bereiter, 1993]. More specifically, the main goal of CSCL is to maximize individual learning based on group work. Recently many CSCL tools such as e-NOTEBOOK system have been developed and embedded in e-learning environments.. CoVis Collaboratory Notebook. The CoVis Project explores issues of scaling, diversity, and sustainability as the relation to the use of networking technologies to enable high school students to work in collaboration with remote students, teachers, and scientists [Edelson and O’Neill, 1994]. And it has appropriated and adapted tools developed for use in scientific and corporate workplaces. The Collaboratory Notebook is a hypermedia tool that supports group project work in science utilizing a shared hypermedia database designed to provide a scaffold for students as they learn to conduct collaborative, open-ended investigations [Edelson and O’Neill, 1994]. By using CoVis Collaboratory Notebook, learners create pages that may be linked together through hypermedia links indicating the semantic relationships between them. Scientific reasoning is expanded via specific link types that are used to construct arguments and record the inquiry process.. CSILE (computer supported intentional learning environments). CSILE is a communal database system in which learners are allowed to externalize their thoughts mainly in the form of texts or/and graphics called “notes”, then engage in collaboratively organizing their knowledge as objects to.

(36) 22. Chapter 2: Related Works. advance their communal understanding as a whole [Hewitt and Webb, 1992]. This communal database structure has been found to provide learners with opportunities to be involved in knowledge advancement through distribution of their expertise, and to eventually facilitate learners’ conceptual understanding of complex scientific phenomena in comparison with traditional instructions. The CSILE is intended to support interplay of private and public reflection through its communal student generated database and commenting functions and it is a powerful tool to transform learning activities to knowledge building.. SenseMaker KIE. SenseMaker is a software component of the Knowledge Integration Environment (KIE). KIE is a cohesive set of software tools and a project-based framework that is focused around Web resources [Bell, 1997]. Students engage with Web resources as pieces of scientific evidence to be interpreted, explored, and applied to their science projects. SenseMaker allows students to construct and edit their arguments using a graphical representation. They review evidence from the web, each item being represented by a dot in the SenseMake argument. Students then make their thinking visible by describing and grouping the evidence using frames. They can add new frames within existing frames or outside the existing frames. Evidence for arguments is represented with a dot and a link to its internet location. It serves as a knowledge representation tool for the students and the teachers by making three distinct forms of thinking visible: modeling expert thinking, providing a sense-making process to support individual reflection, promoting the collaborative exchange of ideas. SenseMaker helps students by making the process of organizing evidence into claims visible.. ReCoNote. ReCoNote (Reflective Collaboration Note) is a note-sharing system with a mutual-linking capability in which students explicitly create relations between notes [Masukawa, 2004]. When learners encounter two “link-able” pieces of information, they are explicitly asked to link them together with specific comments regarding the relation. The comments are stored and presented in a list whenever the attached notes are viewed. The system’s mutual linking feature requires the learners to adopt multiple perspectives on the learning resources by asking them to write link comments bi-directionally..

(37) Chapter 2: Related Works. 23. Collaborative learning is supported in above mentioned systems. Learners are able to express their ideas, record their activities, and communicate with others by using these systems. However, with the increase in numbers of learning resources (web pages, notes and comments) and learners, following limitations prevent effective and efficient group learning in e-NOTEBOOK system: •. The difficulty in finding relevant contributions and suitable learning resources (e.g. web pages, notes and comments).. •. The difficulty in finding learners to collaborate with.. Proposed e-NOTEBOOK system allows overcoming these limitations by implementing the functions such as recommendation of learning resources and recommendation of advanced learners.. 2.2 Recommender system. In this section, firstly, we begin with the general recommender system model and describe the main techniques of recommender system, and then give some examples. Secondly, we give a brief comparison of recommender system for e-learning and for other domains (e.g. e-commerce) and formally state the problem of recommender system for e-learning. Thirdly, we discuss the related works on content filtering based on keyword map. Finally, we discuss the related works on collaborative filtering based on learner’s relationship.. 2.2.1 Framework of recommender system Recommender systems [Resnick and Varian, 1997] are typical examples of personalization systems and typically suggest items (e.g., products, documents) that are of interest to users, according to user demographics, features of items, and/or user preferences and provides users with recommendations about products and services they may like. Generally, the goals of a recommender system can be summarized as follows [Chaptini, 2005]: (i) Delivering relevant recommendations based on each individual’s tastes and preferences; (ii) Determining these preferences with minimal involvement from users; (iii) Delivering recommendations in real time, enabling users to act on them immediately. Therefore, recommender systems work by collecting data from users, using explicit and/or implicit methods and then comparing the collected data to the active user data to find a list of.

(38) 24. Chapter 2: Related Works. Fig.2.1 Framework of a recommender system. recommendations for the active user. Accordingly, the three modules involved in a recommender system include: (i) Input sources; (ii) Recommending method; (iii) Output recommendations [Schafer et al., 1999] [Montaner et al., 2003] [Wang and Wu, 2009]. Fig.2.1 is the framework of a recommender system. In the follows, we will briefly review the current developments with respective to these three modules.. 1) Input sources Input sources usually consist of two parts: user profiles and reference characteristics. In general, there are three types of the reference characteristics: information from the social environment, namely the social environment provides characteristics; information about the items themselves, that is, item provides content-based characteristics; and information about web usage, in other words web usage provides characteristics. On the other hand, the input sources of the user profiles include the user’s preference from the specific items, preferred item attributes, ratings, and keywords or even purchase history [Schafer et al., 2001]. And, the user profile information can be elicited from demographical data, user preferences about features of the items, and user ratings on experienced items [Claypool et al., 2001].. User profiling is the process of gathering information specific to each user. It is typically either knowledge-based or behavior-based and can be accomplished either explicitly or implicitly. Knowledge-based approaches usually use questionnaires and interviews to obtain user knowledge. Typically, the system offers the user the opportunity to rate any item on a discrete scale of 5 or 10.

(39) Chapter 2: Related Works. 25. grades. For example, GroupLens [Resnick et al., 1994] acquires user profiles by asking users to grade the pages they have browsed for interest and relevance by using explicit method. Behavior-based approaches use the user’s behavior as a model, and discover useful patterns in the behavior by using machine learning techniques. Behavioral logging is made use of obtaining the data necessary from which to extract patterns commonly. This method acquires user profile by estimating the users’ degree of interest in the items the users have accessed based on such factors as (i) the time spent reading the items (reading time) [Wan et al., 2007, 2008] [Morita and Shinoda, 1994] or (ii) the specific mouse button operations or the scroll operations performed while reading the items [Goecks and Shavlik, 2000], or (iii) the user’s eye mark while reading pages [Ohno, 2000], and etc. by using implicit methods.. 2) Recommending methods Essentially all the recommender systems have the same aim: to lead users through recommendations to those products which are the most suitable for them. However, the techniques utilized to achieve this aim are different from each other, so in the required information as in the necessary processes to compute the recommendations. For a recommender system, it is critical to find out the user’s purchasing behavior in a systematic way so that it can support the decision of item (such as products, documents) selection. The recommending method is a module which serves this purpose of item selection. Three types of recommending methods are commonly implemented [Adomavicius and Tuzhilin, 2001]: Content-based filtering (CBF), Collaborative filtering (CF), and Hybrid approaches, based on how recommend actions are made. These methods have, however, their inherent strengths and weaknesses. The recommender system designer must select which strategy is most appropriate given a particular problem.. 3) Output recommendations Usually, the output turns out to be a suggestion with the information of item type, quantity and appearance [Schafer et al., 1999]. The output of a recommender system is generally a set of items with the highest predicted interest values or utilities for an active user. The way a recommendation is presented may also show how good or relevant the item is considered to be. Relevance can be.