Architecture of Web-Based Intelligent Collaborative Learning : System Design and Peer Modeling

Article excerpt

This paper discusses architecture of web-based intelligent collaborative learning system (WebICL) and delves upon (systems design and peer modeling). A systems framework has been introduced, which has been designed based on the web-based collaborative learning fundamental theory. The peer modeling includes knowledge-based peer modeling, general peer model and virtual peer model in WebICL. The objective of this study is to implement an intelligent and flexible web-based intelligent collaborative learning system and to facilitate learners learning performance in web-based environment.

(ProQuest-CSA LLC: ... denotes formulae omitted.)


Collaborative learning approach has dealt primarily with standard, classroom-based environment and not the web-based environment, which raised the question of how well the benefits of collaborative learning will translate to the web-based environment (Brandon, 1999). Collaborative learning means that knowledge is not something that is 'delivered' to students, but rather something that emerges from active dialogue among those who seek to understand and apply concepts and techniques. Student-student interaction in collaborative learning communities may contribute to the achievement of educational goals by influencing educational motivation and aspirations through peer relationships (Hiltz, 1993; Johnson, 1981). To collaborate means to work together, which implies a concept of shared goals, and an explicit intention of "add value"-to create something new or different through a deliberate and structured collaborative process, as opposed to simply exchanging information or passing on instructions (Kaye, 1994). Collaborative learning would be the acquisition by individuals of knowledge, skills, or attitudes occurring as a result of group interaction, or put more tersely, individual learning as a result of group process (Kaye A, 1992). Most outstanding experiments show that group learning is an important element of collaborative learning. Group learning is based on making individual fell responsible towards the group (Trentin, 1999). The idea that collaborative learning is the development of shared meaning among group members reflects the larger CSCL (computer supported collaborative learning) perspective on learning, a perspective that emphasizes the social creation of knowledge as the basis of learning. Meaning is not pre-packaged and delivered to the student for memorization; rather, it is negotiated among group members (Pea, 1994; Roschelle, 1992).

According to the above description, the main important components of collaborative learning in web-based environment is how to realize social context, group learning process, communication with each other (collaboration), and performance evaluation. Web-based collaborative learning is not only computer programming but also system integrated collaborative learning which depends process & social context. It should be more flexible and suitable for on-line group learning (Brandon, 1999). The collaborative development of requires a substantial amount of communication, perhaps even more so in on-line than in face-to-face groups. The intelligent learning method needs to be integrated into web based collaborative earning system because the intelligent characteristic is the basis of otherfeatures, suitable, practical, available.


Web-based collaborative learning system can be divided into two categoriesone is asynchronous system and another is synchronous underwhich many practical systems were developed. The popular asynchronous system includes First Class, CSILE/Knowledge Forum, Learning Space, WebBoard, and WebCT; synchronous system includes Conference MOOS, WebChat Broadcasting System and Microsoft Netmeeting.

First Class is a very practical conferencing system accessible through a browser. Its features include 'software which has a rich environment that offers both real-time (synchronous) facilities and delayed time (asynchronous) resources. They include chat, shared documents (can be created and edited amongst a group of people), bulletin board facility, mail, and conference and discussion groups etc (McConnell, 2000). CSILE/Knowledge Forum was developed by a team of cognitive research scientists in Toronto and teachers across Canada (Scardamalia et al, 1989). Typical notes in Knowledge Forum include a question, a problem, a graphic illustration, a research plan or a summary of information found from resource material (Simons, 1999). CSILE offers an open learning system where students and teachers can share their knowledge and work collaboratively in building new knowledge notes. Each database is open to all Knowledge Forum users. This model of learning supports a constructivist approach to knowledge (Scardamalia, 1999). Learning space is a course-authoring environment for web-based teaching and learning. It includes both synchronous and asynchronous facilities as well as group and individual spaces. There are a lot of differentfacilities and can also contain documents in many different formats. Links to other parts of the course environment and beyond are also easily included. The features of learning space include CourseRoom, Schedule, MediaCenter, Profiles, Assessment Manager, and Learning Server. WebBoard is a web-based system that can run as its own server. Discussions are structured into Forums and threads. Features of this system include chat facility (real-time discussion), conferencing facility (threaded discussions amongst a group), attach files to messages, and web server functionality. WebCT is a web-based authoring and electronic communications system developed at the University of British Columbia. There is a bulletin board system, which allows users to discuss matters of interest and post information to each other. The feathers of WebCT include web-based tools (authoring pages of text, graphics, etc), Chat (real-time discussions that can be logged), email facilities for individuals (as well as group mailing list facilities), and conferencing facilities (threaded discussions) (McConnell, 2000). Conference MOOs are Multiple Object-Oriented systems which are text based. Many users can link into a MOO and communicate at the same time, using text dialogue boxes divided into classrooms, hallways and other virtual meeting places. Linking many chat systems, users can prepare their text in a window without others seeing it before sending it. Once sent, it appears in the shared chat history. Students can also carry out other actions, such as waving, smiling and so on, through recognized non-verbal gestures. MOOs are sometimes used to support decision making and other tasks that benefit from synchronous communication. Diversity University MOO is one of the most well-known of these environments tailored for educational purpose. It is organized around a campus metaphor, where in features include writing tools, communication tools, exploration commands, manipulation commands, bulletin board, videoconferencing handler, VRML view, ghost view, and MOO map. WebChat Broadcasting system allows the user to communicate in realtime envi ronmentwith other users via text-based messages. Recent versions of the application also support the uploading of images during chat. In addition to supporting chat, the home web site incorporates a number of other features which support the community orientation of its service including: the ability to search the profiles of those currently online; setting up your own homepage; a news stand; instructions on establishing a Net Circle; and an electronic mail account. Since its arrival in 1993 WebChat has attracted overS million users. In 1998, WebChatwas acquired by the Internet service lnfoseek. Its features include browser-based organization into topic forums, real-time discussion, different chat models, steaming chat, frames, no frames, allows uploading of images during chat, and allows private chat, etc. Microsoft NetMeeting is Windows-based collaboration tool incorporating data, and video conferencing in one package. Its features include collaboration through windows-based applications, data conferencing, electronic whiteboard, file transfer, text-based chat, audioconferencing, videoconferencing, and support communication with users using compatible products (McConnell, 2000).

Intelligent instructional method can be used into collaborative learning electronic system to enhance its efficiency and flexibility. Some promising issues in this field have been addressed, e.g., intelligent pedagogical agent, student model, tutor model, diagnosis strategies, knowledge intelligent representation, etc. The important issue while applying intelligent method to collaborative learning system is to find its integrated position and approach based on collaborative learn ing features. CITS (Collaborative Intelligent Tutoring System) provides an environmentwhere the student can interact with one or more, simulated collaborative partners and/or fellow students, to progress towards a common goal of learning (Kumar, 1992). GSS (Group Support Systems) is a set of techniques, software and technology designed to focus and enhance the communication, deliberations, and decision making of groups. In GSS, software intelligent agent can facilitate and streamline group problem solving in organizations which is applied (Nunamaker Jr, 1997; Sen, et al, 1997). Brna& Burton (1997) described modeling students collaborating while learning about energy, which has the potential for providing better computer-based support in the future - both in respect of providing improved quality dialogues and in terms of comprehending the student's activities. Miyahara & Okamoto (1998) studied how to develop an information filtering system, which gathers, classifies, stores various kinds of information found on the Internet.

Our work is related to what is described above, which is integrating collaborative learning and intelligent method together to construct one flexible and adaptive intelligent collaborative learning system. According to our study, intelligent method is very important to enhance the quality of web-based collaborative learning, e.g., flexible, efficient, and suitable, etc. and to facilitate students learning performance. We built peer model and integrated itintoourleamingsystem-WeblCL. Learner can easily immerse with online collaborative learning process because WeblCL is based on collaborative learning procedures. In this paper, we described how to design system framework and how to simulate peer model. Other related issues, e.g., peer modeling method, evaluation method and collaborative learning factor analysis, etc. were introduced in other papers (Zhao, Li & Akahori, 2001a; Zhao & Li, 2001b; Li & Zhao, 2001).

WeblCL System Modeling

To use WeblCL as an instructional system, system analysis and design techniques should be utilized first. Instructional system design can be adopted as a good approach to realize it. The framework of web-based intelligent collaborative learning system will be designed based on the following phases:


According to Slavin (1995), the goal of collaborative learning is for student to help each other succeed academically. To be successful, all members in a group must achieve mastery of the material or contribute to the completion of a group assignment. Theoretically, collaborative learning fosters a cooperative atmosphere in classrooms, ratherthan a competitive one, because students are investing in each other's learning, not just their own. Edelson et al (1995) stated that student participation in collaborative, open-ended inquiry is a central goal of many current science education reform efforts. McManus (1996) indicated that several goals of cooperative group learning have been identified in the literature. Two primary goals for all students are (a) to assume leadership responsibilities in the group, and (b) to participate equally and actively in the group process. Additional goals of collaborative learning include fostering academic cooperation among students, encouraging positive group interaction, increasing academic achievement, and developing self-esteem.

The main objective of WeblCL is to design and implement a flexible and integrated collaborative learning system to facilitate student learning performance in web-based learning environment, a mannerthat classroombased collaborative learning and intelligent methods can be integrated together. The objective of WeblCL can be divided into three categories collaborative leaning objective, learning group's objective, and peer's learning objective, lnaba et al, (2000) described how to use learning goal ontology to form effective learning groups. Their perspective is useful to find a new way to clarify and define the objective of WeblCL.


Feasibility of WeblCL system mainly indicates instructional and learning requirement. It can be divided into three categories systematic requirement, tutor's requirement, and peer's requirement. Feasibility of WeblCL system provides the foundation of system design and modeling, which the precondition to develop a flexible and adaptive web-based learning system.


Systematic requirement includes development, implementation, operation, and interaction, etc, which will impact flexibility and adaptability of WeblCL. Systematic requirement can be modelled according to tutor's requirement and peer's requirement. The outcome of systematic requirement will be the basis to design the system framework.


Within WeblCL environment, the role of the tutor is in most respects no different to their role in face-to-face cooperative learning situation (McConnell, 2000). Tutor's requirement includes how to facilitate teaching how to organize instructional approach and how to realize the teacher's role of leader, designer, facilitator, guider, assistant, evaluatorand assessor.


Peer's requirement includes learning content (curriculum knowledge), learning resource, interaction approach, learning tools, learning environment, and systematic interface. The objective of WeblCL system is to provide a flexible and adaptive on-line learning environment to facilitate student learning performance. Peer's requirement should become the main focus that WeblCL system will service. Cognitive learning and other related theory can be used for this process.


According to McConnell's (2000) experiences, useful and important aspect of CSCL design includes openness in the educational process-the learning community, self-determined learning, a real purpose in the cooperative process, a supportive learning environment, a collaborative assessment of learning and assessment and evaluation of the ongoing learning process.

The strategies and principles which can be used in system design include objective determining, research area and object, system function, investigation about user, control and monitor strategies, and evaluation strategies.


There are eight modules in WeblCL system, which are peer module, group module, interface module (peer and tutor), database module, curriculum knowledge module, evaluation module, tutor module, and CLtools module. The working mechanism and processes of each module is described as follows.


When a student logs in through student interface, WeblCL system will search his register account number in student records database. If his account is found, it will be used to search the learning history records from student models database. Then the data of student model is acquired and sent to student grouping module. If student's account number cannot be found in student records database, a new account number will be appended in the student records database when the student finishes his register form. Generally, when a new student registers in WeblCL, he will be asked to participate in the pretest or psychological survey based on his knowledge background. The result of pretest or/and psychological survey and personal messages that come from registerform will be used to form student model. Student learning history records or his new learning records are called nature data (ND), which can be used to form student model and group model. This process can be simply described as Figure 1.


Organizing learning group process includes two statuses. The first status is that the data of learning group structure will be fetched from group structures database based on ND from learning history records. This data can be exchanged with student grouping module. ND from learning history records can be sent to student models database. Student model will be formed based on ND. There are many strategies and principles in student grouping module which can be used to organize learning group based on student model.

Figure 2: Organizing Learning Group Process

The second status is related new learning record. Student model will be simulated in student models database based on ND. The data of student model can be sent to student group module, which can be transmitted to group structures database. Then it can be organized into a group in term of grouping strategies and principles. If the number of students who login the system is smaller than necessary of a learning group or the knowledge background is not suitable to the strategies and principles of group learning then WeblCL can simulate virtual student model based on the necessary requirements of group learning and organize it into specified learning group with online student. The organizing learning group process can be described as in Figure 2.


Learning Knowledge in WeblCL includes two knowledge databases. One is learning task database and another is learning resource database. The form of the learning task in WeblCL is problem-based knowledge, which can be divided into two categories. One is curriculum sequence knowledge and another is integrative curriculum knowledge. Learning resource includes various knowledge backgrounds, picture, graphic, audio, essay, video, and animation, etc, which are multimedia styles related learning tasks. Learning contents are saved in the learning task database, which can be presented on knowledge presentation module. The presented contents to learning group can be selected based on the ND from student records database and group structure database in learning tasks database. Student in learning group can freely get learning resource based on the learning task resolving. The process of learning task presentation can be described as Figure 3.


Teacher's role in WebICL is most similarwith face-to-face collaborative learning environment. The Teacher is one of the most important elements in web-based collaborative learning environment. When Teacher (tutor) logsin the system, he can control and monitor collaborative learning process through tutor interface. Four service modules are provided to teacher in WeblCL, which are student model service modules, group structure service module, virtual teacher service module, and knowledge database service module. A teacher can delete or append student model, group structure, virtual tutor, and learning material through tutor interface, which can modify the data in these four services modules. When teacher logsin the system through tutor interface, he/she also can join the collaborative learning process. Tutor model can be established based on the teacher's personal message, academic speciality, and experiences. Virtual tutor in WebICL undertakes real teacher's task, which will be that of a facilitator, leader, designer, guider, mediator, assistant, evaluatorand assessor for student collaborative learning performance. In WebICL, virtual tutor is regarded as tutor agent, which can act like a real teacher. Tutor or virtual tutor also can control knowledge database to present the specific content to learning group or specific student. He also can assign a student to specific learning group. Figure 4 describes the teacher's role in WebICL.


The achievement of web-based collaborative learning is embodied in essay, report or presentation (home page or PowerPoint production, etc), which the outcomes of learning group and the principal part for evaluation. Evaluation in WebICL for peer collaborative learning is more concerned with collaborative learning outcomes and process.

The process of evaluation for collaborative learning outcomes includes two levels. One is learning group level and another is peer individual level. In the learning group level, when learning group finishes one task, teacher (tutor) and other learning groups will assess their outcome and assign one score and remark to it. This score is normally called learning group score. In peer individual level, the learning group score will be averaged to each peer mate in learning group. This is peer's basic score for his collaborative learning.

The process of evaluation for collaborative learning process can be divided into three categories. The first is for learning knowledge presentation. The results of evaluation for learning group can be used to control the learning material in knowledge database to groupmates based on the learning group score and discuss, collaborate, debate and conflict etc and receivce feedback messages. There are some deducing strategies and principles which are saved on the deducing mechanism. The second category is for peer individual learning performance, which is very important for how to determine the peers' learning performance. Evaluation for peer individual learning performance can be done by tutor and peer mates in the same learning group. The tutor assigns his assessment score based on his observation (not for every peer mate) and peer mates give his peer mates (except himself) learning score based on his feeling. The main criterion of evaluation is one's contribution to his learning group. This score can be added on the peer individual's basic score. The third category is related to how to save the evaluation data. When peer in learning group logsout from WebICL or during the collaborative learning process, the evaluation data will be saved in the student records database. This data can be used by student models database to form peer model and to organize them into specific learning group. The process of evaluation in WeblCL can be described as in Figure 5.


Peers in collaborative learning groups can use collaborative learning tools to communicate with his learning partners. These collaborative learning tools can be divided into two categories. One is asynchronous learning tool which includes email, presentation tools, bulletin boards, and search engine, etc and another is synchronous learning tools which that it includes internet phone, chat room, video conference, and seminar room etc. These learning tools can be used to facilitate and to enhance peer's learning performance. The categories of collaborative learning tools can be described as Figure 6.


According to the description above, the systematic framework of WebICL can be designed as Figure 7.


Peer model is a core element of the small learning group to construct the web-based collaborative learning environment, which can be treated as a virtual real learner. The differentiation of peer model can be found in webbased collaborative learning environment and in an intelligent tutoring system (ITS) Its purpose in ITS environment is to facilitate individual learning performance and in web-based collaborative learning environment the purpose is to promote individual and collectivity development and to enhance peer interaction. The term peermodel comes from student model, which are similar concepts. When students are organized into different learning teams, they are called peers. The prerequisite to understanding peer model is to understand the student model.


Student model is typically used in connection with applications of computerbased intelligent instructional systems. In this context, the student model is a representation of the computer system's belief about the learner and is, therefore, an abstract representation of the learner in the system. There are various ways to reason about the learner's knowledge. If one keeps a simple history of learner behaviour as a data source, drawing inferences about the subsequent inference is easier. If one first makes interpretations of the learner's behaviour, subsequent inference is easier. Thus one could construct explanations of behavior in light of prior explanations and save the new explanation (Holt; Dubs; Jones, and Greer, 1991). Student models have been devised that record either misconceptions, missing conceptions or a combination of both. There are overlay student models, differential student models, and perturbation or buggy student models (Smith, 1998).


According to the description of student model above, the core question to build peer model is how to monitor the learner's learning behavior and evaluate the peer's performance in learning group. There are various practical and efficient approaches about peer modeling in web-based environment. For example, Kumarefa/, (1995) observed artificial intelligent techniques in CALL and discussed techniques that had been used for diagnosis and the representation of student models. Eliot (1997) constructed a web-based intelligent tutor in which system requires information about what student has and has not learned, in order to focus on educational activities that will be most beneficial to the student. Although this tutoring system does not include collaborative features, but the technological foundation could easily support collaboration. Stern (1997) stated that difficulties in web-based tutoring and some possible solutions. According to his opinion, a tutoring system must record student actions and make decisions based on those actions. Oliveira (1996) suggested the intelligent agent could be used to realize the student modelling in web-based CL. He also stated the importance of goals and said, "each learner interacts with his/her assistant in order to define goals. Once the goals are defined, either the learner or the assistant may suggest a plan to achieve it. The agent can accept the plan, reject it, or suggest an alternative, and so on. The assistant can also ask the other assistants for suggestions. All of these interactions follow a single cooperative learning protocol, which is a variation on Koning's protocol (Koning, 1995)". In this paper, one knowledge-based peer model will be described, which is easier to apply than those described above. Peer model includes a lot of factors, e.g., knowledge structure, cognitive styles, behavioral factors, and physiological elements, etc. The best and natural peer model should include all of these factors, which impact the peer's learning. One of the most important factors in these elements, which can be considered experts knowledge structure, is knowledge point. This transformation method can reduce the difficulty and complexity that we encounter in peer modeling. Our previous works focused on the construction of knowledge-based student model (Zhao, 1997). This resolution can also be considered as a general approach to build the peer model in web-based collaborative learning system. The process of knowledge-based peer model will be described as follows.

Intelligent technology can be used for collaborative learning electronic system to enhance its efficiency and flexibility. Some promising issues that have to the addressed, are intelligent pedagogical agent, student models, tutor models, diagnosis strategies, knowledge intelligent representation etc. The important issue while applying intelligent method to collaborative learning system is to find its integrated position and approach based on collaborative learning features. CITS (Collaborative Intelligent Tutoring System) provides an environment where the student can interact with one or more, simulated collaborative partners and/orfellow students, to progress towards a common goal of learning (Kumar, 1992). Group Support Systems (GSS) is a set of techniques, software and technology designed to focus and enhance the communication, deliberations and decision making of groups. In GSS, software intelligent agent that can facilitate and streamline group problem solving in organizations is applied (Nunamaker Jr, 1997; Sen, etal, 1997). Bma, and Burton (1997) modelled students collaborating while learning about energy, which has the potential for providing better computer-based support in the future - both in respect of providing improved quality dialogues and in terms of comprehending the student's activities. Miyahara & Okamoto (1998) studied how to develop an information filtering system, which gathers, classifies, stores various kinds of information found on the Internet. According to our experiences, intelligent technology can be used to enhance the quality of web-based collaborative learning, e.g., flexible, efficient, and suitable, etc. and to facilitate students learning. Peer model is the core component to apply intelligent technology into instructional system. In web-based intelligent collaborative learning system (WebICL), peer model can be used to adjust the related instructional strategies and to precisely evaluate peer performance.


Knowledge-based peer model means how to resolve and integrate learning knowledge into intelligent tutoring system to form peer model. In intelligent tutoring system, the behavior of peer includes a lot of knowledge processing variables, which can be effectively used to construct peer model. We will introduce an approach to realize knowledge-based peer model, which is a common method and has drawn from a special intelligent tutoring system for Chinese language learning (Zhao, 1997). The process of building knowledge-based peer model can be divided into four steps.


In web-based collaborative learning system, the core factors must be found and integrated into peer modeling process, that is, these variables can be controlled and operated. One approach to realize this objective is to determine the relationship among these factors with mathematic calculating formula. Some useful methods can be used to do it, e.g., literature research (meta-analysis), questionnaire, action research, and case study. Then, an evaluation system for these factors should be constructed. In the end, the relationship among these factors can be formulated utilizing mathematical method. According to our experience, questionnaire survey is a good approach to collect related data. Teachers and instructional experts are the object of our questionnaire survey. The items in questionnaire should be evaluated several times before the formal survey is carried out.

According to our questionnaire survey result, core factors of web-based collaborative learning system can be divided into four levels, the first is learning approach level, which includes group composition, task features, communication media, conflict identification, conflict resolution, role playing and collaborative communication skills (Foote, 1996); the second is personal message level, which includes gender, age, region, grade, ethnic and year-in-school, etc; the third is peer's cognitive structure level, which includes cognitive styles, behavior styles, previous academic performance, and IQ, etc; the fourth is curriculum knowledge level, which includes knowledge structure, knowledge points, hierarchy of knowledge points and knowledge categories etc. These four levels of core factors of web-based collaborative learning system can be listed in Table 1.

The core factors of these four levels can be used to build peer model. As described above, learning knowledge is more useful to build peer model because it is the umbrella element to resolve peer modeling question (Zhao 1997). We will now introduce how to realize a knowledge-based student model in intelligent Chinese language learning system.


One effective and adaptive approach was found to build knowledge-based peer model based on ourformer research. Thefoundation of this approach can be elicited from the core factors of course content by questionnaire and factor-analysis method. Then assign the weights to each factor by questionnaire survey and data analysis method. This approach can be used as a common method and applied to any subject (Zhao, 1997).

According to Dr. Mo lei's research about Chinese reading (1989), the Chinese reading ability of student in primary Grade six in China is listed in Table 2. As we all know, students in different grade will face a different instructional knowledge structure. If we address Chinese reading student model in primary school in Grade three, the data in Table 2 must be adjusted to adapt to Grade three.

The result we elicited was listed in Table 3. Item of instructional content and the weight of each factors of Chinese reading ability are included in the Table 3. According to the factors' weight in Table 3, the basic knowledge of peer model and the dynamic peer model can be built.


Based on the factors' weight in Table 3, we can formulate the basic static knowledge peer model as follows,

Where G is the objective weight matrix.

When peer registers to the WeblCL system, he/she will be provided the pretest opportunity. The score of each learning objective will be recorded and one pretest evaluation matrix will be formulated as follows,

P = (S^sub 1^ S^sub 2^ S^sub 3^ S^sub 4^ S^sub 5^ S^sub 6^)

Where P is the score matrix for peer and s is score that he/she got from each question; st is the score for each learning objective; i = 1,2,3,...6.

Following regression to P,

P = (s^sub 1^/t^sub 1^ s^sub 2^/t^sub 2^ s^sub 3^/t^sub 3^ s^sub 4^/t^sub 4^ s^sub 5^/t^sub 5^ S^sub 6^/t^sub 6^)

Where t is the mark of each question; t = 1,2,3,...6.

We can get the evaluation result for to peer through fuzzy evaluation methods.


Where A is the result of evaluating a student.


Dynamic knowledge peer model can be simulated based on the basic static knowledge peer model in WebICL system. Tutor and system will monitor peer's action when peermates start to discuss or negotiate each other in learning group. The system can record and save the group and peer's learning history record. When they finish one topic, tutor will assign the average score to each peer in the same learning group based on the result of intergroup evaluation or other similar approaches. To get the peer's score and get the exactly peer performance score, peer should participat in the individual formative examination. When the peers are taking a formative examination, two elements will be refered, which are peer's performance score and the time of peer resolving question which will influence peer's performance score. To get this score about student resolving question, the following strategies will be used (Zhao, 1997).

Where times refers to number of peer try times to solve the question; score is each question's marks; Score_i is the student's factual score of each question.

The time which peer resolving each question spends also influences the peer-performance's score. The following strategies can be adopted (Zhao, 1997).


Where spend is the average response time (time limitation for problemsolving); spent is the factual time which peer spends to resolve question; times=1,2,3...n.

The data of static knowledge peer model will be modified when peer finishes the formative examination. Then the dynamic knowledge peer model will be built based on the result of fuzzy evaluation method. Each item in examination will be formed based on the learning objective by tutor. Peer can get the different evaluation scores when he/she answers the different questions and one score matrix will be obtained. This matrix B can be described as follows.


Where B is score matrix of peer: s^sub ij^ is each score for different objective;

i = 0,2,3,...6; j = 1,2,3,...6.

Peer knowledge dynamic model will be built by the formula as follows. This formula indicates the peer's knowledge mastering situation for each learning objective.

... (2)

Where A is result of fuzzy evaluation which represents the dynamic knowledge peer model.

The main feature of dynamic knowledge peer model is that its data will be constantly modified and updated according to the learning objective. Web-based collaborative learning system can realize the suitable and flexible objective according to dynamic knowledge peer model. The score of dynamic peer model can be considered as one of the components of peer's learning performance with average score of learning group performance as assigned by tutor.


Peer model in WebICL includes two categories. One is general peer model and another is virtual peer model which can also called peer agent. There are two kinds of peer model.


General peer model in WebICL will represent real student who will register (login) in the system. Traditionally, there are following basic elements in general peer model, which are gender, former learning performance, IQ, and cognitive style. The general peer model can be simulated based on these elements, which can be described as a one-dimension at matrix.

GPM= (iG A iQ iCS)

Where GPM means general peer model, iG means gender, A means student learning performance which was introduced in the above formula (1) and (2), iQ means student's IQ, and iCS means cognitive style.

The value of IQ and cognitive style can be evaluated before collaborative learning in web-based environment through psychological survey. A is the result of fuzzy evaluation to student.

Three elements are the permanent variables, which they are iG, iQ, and ICS. The result of iG can be got by peer's personal registered form; iQ and iCS can be got by psychological survey. The result of iCS is mainly used in dependent and independent style field survey, which can be got before peers starts collaborative learning process.


Virtual peer model (VPM) can be organized into the same group with general peer model if the number of peers cannot reach the group size. VPM includes three components, which are knowledge database, strategies database, and deducing mechanism. The virtual peer model can be described as Figure 7


Knowledge database provides knowledge support for virtual peer agent, which is similar to knowledge database in WeblCLsystem. It is more related to learning task. The structure of knowledge database for peer agent includes five elements, which are knowledge number, knowledge types, knowledge content, question answer grade level. Knowledge number means knowledge record number in knowledge database. Knowledge types means the types of discipline and task, knowledge content uses to record learning content. Question and answer records learning question and its normal answer, and grade level represents which grade level of learning task.

KD = (Kn Kt Kc Qa Gl)

Where Kn represents knowledge number, Kt means knowledge types, Kc is knowledge content, Qa means question and answer, and Gl is grade level.

Kt = (a b c)

Where a = 1, 2, ..., n means discipline, b = 1, 2, ..., n means task, and c = 1, 2, ..., n means difficult level.

Where d = 1, 2, ..., n means content number, e = 1, 2, ..., n means content index number, and f = 1, 2, ..., n means learning content showed times.

Qa = (g h i j k)

Where g = 1, 2, ..., n means the number of question, h = 1, 2, ..., n means the number of answer, i = 1, 2, ..., n means the difficult level of question, j = 1, 2, ..., n means the question showed times, and k = 1, 2, ..., n means the index number of question.

Gl = (m q)

Where m = 1, 2, ..., n means educational level (primary school, junior school, senior school, etc), and q = 1, 2, .. n means grade level.

Teacher can add or delete knowledge item in knowledge database based on the necessary of collaborative learning.


Relevant strategies are important for how to establish the virtual peer model based on the on-line peer's situation in the WebICL system and organize them into situated learning group. There are five elements which are included in the strategies, which they are strategy index number, performance level, role-playing, peer number, and role gender.

SD = (Si Pl Rp Pn Rg)

Where Si means strategy index number, Pl represents performance level, Rp is role-playing, Pn means peer number, and Rg means the status of role gender.

Si = 1, 2, ..., n

Pi = (Hi Mi Lo)

Where Hi means high level, Mi means middle level, and Lo means lower level.

Rp = 1, 2, ..., 18

The value of Rp represents different role in learning group, which can be described as Table 4 (Forsyth).

Pn = 1, 2, 3, 4

The value of Pn means the peer number in learning group.

Rg = 1, 2

Where Rg = 1 means peer is male and Rg = 2 means peer is female.


Deducing mechanism is representative of expression and discriminant. The virtual peer model will be formed based on its deducing result. The condition-decision is the formal approach to realize the deducing mechanism. The basic decision norm is rules that define the arguments to deduce. There are various rules which can be used to make deducing decision. One of the examples is as follows.

If Then

Case 1 if group_number=2 then

Case 2 if group_number=1 then < determine other two VPM based on the peer's situation>


WebICL system is a practical, flexible, and intelligent system, which provides one platform for an on-line collaborative learning environment. Artificial intelligence can be integrated into WebICL system through system design. WebICL system is designed based on collaborative learning and other related theory, e.g., cognitive learning, situated learning, problem-based learning, constructivisim, and instructional design, etc. There are a lot of approaches. However the knowledge-based peer model method is a practical way and can be effectively applied into WebICL system.

Our research proves that system design is a very important stage if we want to develop online learning system. This process should be refined based on instructional system design process, which is the foundation to guarantee our developed system being an instructional system.

The key element of peer modeling is how to evaluate. In this paper, this issue was addressed. But the prerequisite of peer modeling is to get the basic factors. If peer model is based on knowledge, it must be got from curriculum knowledge. The evaluation method will be designed aiming at these factors.



Inaba A., Thepchai Supnithi, Mitsuru Ikeda, Riichiro Mizoguchi and Jun'ichi Toyoda. (2000) "Learning Goal Ontology to Form Effective Learning Groups, "in proceedings of the International Workshop on New Technologies for Collaborative Learning, Kyoto, Nov 27-29.

Brandon, O.P. and A.B. Hollingshead (1999) "Collaborative Learning and Computer-Supported Groups," Communication Education, 48(2): 109-126.

Brna, P. & M. Burton (1997) "Modelling Students Collaborating While Learning About Energy," Journal of Computer Assisted Learning, Vol. 13, pp. 194-205.

Christopher, Eliot (1997) Implementing Web-Based Intelligent Tutors. Center for Knowledge Communication, Department of Computer Science, University of Massachusetts.

Daniel, C., D. Edelson, Kevin O'Neill, Louis M. Gomezand Laura D'Amico. (1995) A Design for Effective Support of Inquiry and Collaboration, School of Education and Social Policy: Northwestern University.

Foote E. (1996) Collaborative Learning in Community Colleges, Clearinghouse for Community Colleges: ERIC.

De Oliverira, P.M. and R.M. Viccari (1996). Are Learning Systems distributed or Social Systems? Institute de Informica PUCRS and UFRGS.

Hiltz, S.R. (1993) "Correlates of Learning in a Virtual Classroom," International Journal of Man-machine Studies, Vol. 39, pp. 71-98.

Nunamaker Jr., Jay F. (1997) "Future Research in Group Support Systems: Needs, Some Questions and Possible Directions," International Journal of Human Computer Studies, Vol. 47, pp. 357-385.

Johnson, D.W. (1981) "Student-Student Interaction: the Neglected Variable in Education," Educational Research, January, pp. 5-10.

Kaye, A.R. (1992) "Learning Together Apart in A. Kaye (Ed.), Collaborative Learning and Computer Conferencing, Series F, Computer and System Sciences, Vol. 90. Berlin: Springer-Verlag.

Kaye, A.R. (1994) Computer Supported Collaborative Learning. TD-Rivista Di Tecnologie didattiche, Vol. 4, pp. 9-21.

Kumar, VS. (1992) Collaborative Intelligent tutoring System: A Learning Environment, Bombay: National Center for Software Technology.

Kumar, VS., P. Srinivas, KSR Anjaneyulu. (1995) Artificial Intelligence Techniques in CALL: A Review, Knowledge based Computer Systems Project, National Center for Software Technology. Department of Computational Science University of Saskatchewan.

Larson, C.O., D. F. Dansereau, A. O'Donnell, V, Hythecher, J.G. Lambiotte, and T.R. Rocklin (1984) "Verbal Ability and Cooperative Learning," Journal of Reading Behavior, 16, pp. 289-295.

Li, K.D. and J.H. Zhao (2001) "Analysis Method for Compositive Factors of Collaborative Learning in Class room-Based and Web-Based Environment," Proceedings of ICCE2001, Seoul: Korea.

McConnell, D. (2000) Implementing Computer Supported Cooperative Learning, London: Kogan Page.

Mia, K. Stern (1997) "The Difficulties in Web-Based Tutoring, and Some Possible Solutions," Proceedings of the workshop "Intelligent Educational Systems on the World Wide Wab," 8th World Conference of the AIED Society, Japan: Kobe, 18-22 August.

Miyahara, K. and T. Okamoto (1998) "Collaborative Information Filetering in Cooperative Communities," Journal of Computer Assisted Learning, Vol. 14, pp. 100-109.

Lei, Mo (1989) Chinese Reading Research for Student in Elementary School, Guangdong: Guangzhou.

Pea, R.D. (1994) "Seeing What We Build Together: Distributed Multimedia Learning Environments for Transformative Communications," The Journal of the Learning Sciences, Vol. 3, pp. 285-299.

Peter H., D. Shelli, J. Marlene, and G. Jim. (1991) "The State of Student Modelling," Student Modelling: The Key to Individualized Knowledge-Based Instruction, pp. 4-25, Berlin: Springer-Verlag.

Roschelle, J. (1992) "Learning by Collaborating: Convergent Conceptual Change," The Journal of the Learning Sciences, Vol. 2, pp. 235-276.

Trentin, G. (1999) "Network-Based Collaborative Education,' International Journal of Instructional Media, 26(2).

Sandip S., H. Thomas and A. Neeraj. (1997) "Satisfying User Preferences while Negotiating Meetings," International Journal of Human Computer studies, Vol. 47, pp. 407-427.

Scardamalia, M. etal. (1989)"Computer Supported Intentional Learningn Environments," Journal of Educational Computing Research, Vol. 5, pp. 51-68.

Scardamalia, M. (1999) Engaging Students In a Knowledge Society, Paper Presented at the Bi-annual conference of the European Association for Research on Learning and Instruction (EARLI), Gothenburg, August 24-28.

Serengul, S. (1998) Intelligent Tutoring Systems, Middlesex University: School of Computing Science.

Simons, R. (1999) CL-NET: Computer Supported Collaborative Learning Networks in Primary and Secondary Education. Paper presented at the Bi-annual Conference of the European Association for Research on Learning and Instruction (EARLI), Gothenburg, August 24-28.

Slavin, R.E. (1995) Cooperative Learning: Theory, Research, and Practice, MA: Allyn & Bacon.

Susan, M., McManusand G. Maribeth (1996) "Teacher and Student Evaluations of Cooperative Learning and Observed Interactive Behaviors," The Journal of Educational Research, Sep/Oct 1996, 90(1).

Teresa, C. (1998) Supporting Collaborative Writing and ITS Cognitive Tools, University of Paris, Dept. of Cognitive Psychology.

Zhao, J.H. (1997) Chinese Reading Instruction System Based on Student Model for Primary: Design and Implementation, Master Thesis Paper, South China Normal University.

Zhao, J.H., K.D. Li, K. Akahori (2001a) "Evaluation Methods for Student Model in Web-Based Collaborative Learning System," Proceeding of IMPS2001, Osaka.

Zhao, J.H. & K.D. Li (2001b) "Modeling and Design in Web-based Collaborative Learning System," Proceeding of ITHET2001, Kumamoto: Japan.

Webb, E. J.M. Campbell, D. T, Schwartz, R. D., Sechrest, L. and Grove, J. B. (1981). Non-Reactive Measures in the Social Sciences, Boston: Houghton Mifflin.

[Author Affiliation]

Zhao Jianhua & Li Kedong, South China Normal University, Research Institute of Educational Technology, Guangzhou, RR. China Email:,