Printer Friendly

Designing agents for feedback using the documents produced in learning.

This article reports on research, which is based on the premise that the main aim of teaching is to provide appropriate feedback to students as they learn. Human tutors make this process easier by asking students what they are thinking about a topic, then relate the answer to this stage or context of their learning. Hence, the first Computer Supported Learning systems were based on a tutorial question-answer format. Since then research has branched out into Learner Modelling and Intelligent Agents to support learning in more open systems. This article looks at computers emulating mentors who analyse the student's documented activities to provide feedback. The activities are analysed using a methodology that looks at the Human-Computer-Human Interface, and a pattern structure is developed, which is based on an ontology of group learning. Agents are designed and implemented using this structure to analyse synchronous and asynchronous group learning processes and to provide feedback. The ontology used in this research is based on the structure provided by Activity Theory where technology plays the role of mediator in the context of student actions.


Computers have been used to assist learning in many domains. This work is an attempt to develop a process of codifying student learning needs focusing on the documents produced in a group-based project course, into rules for agent support and a structure for learner modelling. The actions of groups of students in an online synchronous system are analysed for patterns, which signify the feedback approach to be taken. The ontology developed by Barros, Verdejo, Read, and Mizoguchi (2002) is used for the pattern structure, based on Activity Theory (Nardi, 1996) format for representing user actions in the learning system.

The subjects are students involved in developing software engineering specifications in groups, which also may work separately as individuals and bring their completed work to the next group session. The logistical overheads involved in communicating, recording and sharing files, and discussion between group members pose a hindrance to learning. In addition, there are the problems of facilitating, tracking, and managing group discussion and outcomes. As a result, without appropriate supervision, it is common for groups to develop specifications that do not meet the course requirements, while clearly they did undertake much design work and development effort. Last, time constraints and locality of group members work against having regular collocated meetings to provide the synergy needed to discuss and review their work.

Based upon these needs, a tool called Intertac-I (Kutay, 2003a) was developed to facilitate storage of files, workflow management, and group communication. The tool enables concurrent planning, editing, drawing, and discussing (Figure 1). This in itself was insufficient as no feedback was provided to further enhance Intertac-I. The documents produced by the system include the project documents, discussion histories, logs of interactions, and agent's rules. In the next version, user modelling data of configuration selections and the history of agent analysis will be added.

The aim of this work is to develop agents which use data mining techniques (Papatheodorou, Vassiliou, & Simon, 2002) on the documents produced to provide appropriate feedback. Intertac-I logs provide the data collection and preprocessing.

This article describes the pattern extraction and development of agents to provide postprocessing in the form of feedback. This feedback is designed to motivate and guide students towards experiences that enable: (a) the generation of the desired conceptions involved in the course; (b) elaboration of these conceptions; and (c) an ability to differentiate between different conceptions. The aim is to enable students to develop a desired depth of understanding and a range of skills in the learning domain. (1)

While focusing on the computer response to student actions, the research methodology places the human interaction in the centre of the analysis with the computer included as a mediating agent, which takes action in response to the history of student interaction. This approach is termed Human-Computer-Human (HCH) as the computer interface has been subjugated to the human's shared interface.

Experiments were conducted with students using Intertac-I, and the results of these case studies were used to develop rule-based agents that are a feature of the second version (Intertac-II). The constructivist principles of learning as enumerated by Savery & Duffy (1995) were used to specify the activity types for agents. These activities include:


1. scaffolding particularly in planning and design;

2. feedback which is instantaneous with programmed text or through e-mail to the tutor;

3. alternative views to provide different approaches and contexts; and

4. reflection support by providing opportunities.

This article is an analysis of the experimental results and a presentation of the pattern language. This language and its underlying ontology for learning and interaction, provide the structure for the implementation of agents in Intertac-II.


Collaborative activity has been posed as having a cognitive advantage in learning through the joint activity (Dillenbourg, Baker, Blaye, & O'Malley, 1996). Collaborative learning has been analysed to understand and interpret the collaborative process to assess the conditions and elements for effective learning (Dillenbourg, Baker, Blaye, & O'Malley). This article explores how some of these learning processes are enacted and develops agents for supporting them, as a method of enhancing learning in online groups.

For this study, students were asked to work in groups of three or four and engage in one of two different specification projects (one easy, the other more complex). Students with little software engineering experience were given the easier project that required them to develop diagrams of a recipe for an Asian meal. The meal involved interdependent courses and various "hardware" implements which could be included or not, depending on how students decided to treat them. Meanwhile students with more software engineering experience were given a more complex task of producing a specification document for a proposed software system described on a commercial web site.

Students were required to use Intertac-I during the course of the experiment. The specifications produced are both textual and diagrammatic in format. While the present Intertac system cannot combine the two into a single document format, the students can use the output to develop a final document in Latex which is a scripting system for document preparation. The data collected using Intertac-I included snapshots of the files at regular intervals plus information on the activities that students were involved in, such as: the opening and closing of tools, contributions to discussion, and the editing changes to diagrams and documents. This data provides a dynamic picture of events or actions.

The data of students' actions is stored in a knowledge base, which forms the aspects of the learning ontology used--configuration elements and historical analysis of process in User Models, processing criteria for the agents in Agent Rule files, the present state of the system in log file. The aspects of each action that can be extracted from Intertac-I log files are: (a) the user initiating the action; (b) the proximity of an object or text to other objects or text; (c) duration or time of the change; (d) conceptual context of the interchange; (e) oppositional forms such as addition and deletions of an object; and (f) user notifying completion of an activity or stage. This knowledge base can be mined for information on the process used by individuals or the group as a whole.

The tools are linked through the Intertac-I system so data is collected that enables agents to:

1. analyse common threads running between the tools;

2. check the timing of document production stages as compared to the planned time-line;

3. compare documents to a template where appropriate;

4. analyse for adherence to rules of the domain; and

5. analyse of interactions and participation within discussion and across all tools.


The research uses a case study approach to document the range of activities undertaken by students. The groups were both self-selected and selected on the basis of various features including: range of experience; years of study; familiarity with domain; and experience in group work. The approach taken is to extract and analyse the activities and tasks in processes used by students in their learning and then identify significant actions in these process.

Since students are working on an ill-formed problem, it is not possible to provide the sort of specific questions, which a computer-based tutorial methodology could be used to analyse. In an alternative approach (Constantino-Gonzalez, Suthers, & Escamilla de los Santos, 2003), students develop public as well as private designs that can be compared to determine advice. In this present work only a single design is developed by the group, or by an individual with the group commenting. Access to students' responses to contributions by others could, in the foreseeable future, come from language recognition tools.

For the present, group interactions and their learning approach are analysed on the basis of repeated processes. Again similar work has been done specifically on interaction patterns in text-based dialogue (Booth, 1992; Constantino-Gonzalez & Suthers, 2001; McManus & Aiken, 1995; Martin, Rodden, Rouncefield, Sommerville, & Viller, 2001), however, these approaches relied on language parsers, which are not used in this work.

There has been much less research done in analysing the patterns in processes used in editing documents and diagrams. (2) The research described here covers patterns of student actions, such as interactions through the dialogue system, work patterns in the interplay between contributions to different applications, and patterns of approaches to learning in the areas of planning, diagrams, and document writing.

On the basis of findings from this research, an Implementation Pattern Language is developed to describe the feedback appropriate for various interactions and learning approaches that were isolated. While these interactions and learning approaches are themselves too disparate to warrant synthesis into a pattern format, the data analysis and implementation of feedback based on these patterns forms a coherent pattern language (Kutay, 2003b) based on a formalism similar to the learning formalism developed by Muhlenbrock, Tewissen, & Hoppe (1998). This approach is similar to that taken in developing patterns to assist in HCI design (Bayle et al.).

Implementation Patterns are necessarily an informal presentation of the concepts and processes of the agents and provide the metadata or ontology of learning in this domain. The ontology is developed in a structure that supports a process of translation from the instructor's definition of the learning objectives and the desired activities and processes of students, to an agent support language. A later section describes the ontology used to translate those requirements into agents.


The experiments were used to specify an ontology for the domain and the synchronous environment. (3) The ontology covers tools and tasks as the context, the source of information as stored in log files and user models and rules for interpretation as used by agents.

The interaction and learning processes that are selected for study are two sources of information on the process: increasing depth of learning; and stage in learning; and two tasks in learning in this domain: efficiency in document production; and efficiency of learning.

The aim of the work is to provide feedback through agents by entering comments in the discussion window or a separate "mentor" window. To develop this feedback, the processes and actions are studied for computer observable patterns. Given the broad scope of interactions that fall under the process and tasks areas previously described, it is necessary to restrict the review to particular aspects that concerned the lecturers in the courses using Intertac-I. The following brief discussion of the results of the investigation into user activity is included to provide an explanation of where the patterns derive from the experimental results. (4)

Depth of Learning

The aim of this section is to provide an analysis of how groups can be distinguished as to their depth of approach to their learning or depth of their conceptions. Some aspects of these differences are only visible if the students choose to use the full attributes of the tools supplied by Intertac, so they may not be able to be analysed in all groups.

One feature of groups working on the projects is the changing focus of discussion and of their work. Even when dealing with a small design, students often talk about other issues, especially as they wait from one person to do some editing. When they do have difficulty with their design they will discuss many issues in quick succession, not fully resolving them. Sometimes they will link these changing foci and build their knowledge progressively, while other groups just appear to search around without direction. Within each project, there can be a limited number of conceptual and data foci for any design and by analysing what the students did in terms of these concepts, a picture emerged of their process for approaching design.

As the students develop or change the way they link or navigate these foci, so does the depth of their learning. The aim is to initiate or develop these links in the group through feedback. The difficulty expressed by students in this experiment and also found by Wood (2001), is the need for this feedback to be pitched at the correct level of understanding, or injected at the correct time in their work.

To track the conceptual development of students, the agents need to provide an accurate way to connect their discussion and actions with each concept of the domain. Intertac-I provides a list of topics in discussion similar to the approach used in CHAOS (Simone, 1994), where they provide a lexicon of terms or jargon from the domain. The lexicon is used to develop agents to support conversational grounding by comparing the history of a user's contributions with the lexicon, and can be extended to include the Data Dictionary as used in E-R representations (Maiden, Cisse, Perez, & Manuel, 1998). The patterns derived from these activities are:

* Lexicon. It is important that the system be able to track the use and development of domain concepts in the learning. From the prior analysis, it is proposed that by providing a summary lexicon, which students can link with their contributions to discussion, students are encouraged to focus their conversation by grounding it in the course.

* Concept extensions. The instructor in their course plan will develop different approaches to the important concepts as examples or questions to ask students. These can be implemented as agents with the aim of encouraging students to link their understanding of a concept to a different focus on the concept.

* Discussion feedback. The students can also receive support for how much and what interchange of conversation has been linked to each lexical word or phrase. For instance has there been much explanation, disagreement, argument, decisions made, and so forth.

Stage of Learning

Another part of the course plan is the timing of the introduction of concepts in the seminars, which are held each week to support the projects, and experience of which aspect of the project the student will be undertaking each week during the session. Again this information can be derived from a course plan by the instructor. This leads to the formulation of the following patterns:

* Template course plan. The timing of interjections in discussion can be assisted by the use of a course plan that provides a time-line for the expected development of various concepts. These are compared to discussion to verify if they are being taken up by the groups.

* Group planning schedule. The course template is available for students to edit. The template provides a guideline or structure for their own group planning and is stored in the knowledge base separately for analysis of stages in the project development. (5)

Example of Process: Approach to Conception of Requirements

The conception of requirement is visible through the changing of requirements; deleting requirements; talking about changes; related changes to DFDs; informing the client (6) of changes to the requirements in the specification; and relating the requirements to the DFDs. In order of depth of conception, the approaches to specification development from requirements can be analysed as follows:

* Null. Few changes to the requirements, and these are not linked to discussion topics.

* Explain change requirements. Changes to requirements are temporally linked to the design of that section of the DFD or B specification. Usually a change in the removal of a requirement. The group often forgets to include reasons for this change for the client. Also changes can be made late in the workshop course (Justify Change Pattern).

* Template course plan. Changes are made during the design stage, and may be linked to an explanation in the document. Changes tend to dimish after this stage, but late changes are frequently a problem for maintaining coherency within the separate components of the grousp documents.

* Context design. Changes to the requirements start in the first session and include some additions as well as deletions. There may be a strong relation between the wording of the requirements and the processes and data of the DFD to trace consistency of late changes, or this link may be provided in the integration section required in the project report.

The main finding that arises from this analysis is that certain processes could be implemented by agents to support deeper conceptions. For example, when requirements are removed, there should be some explanation in the document relating to this removal, or change, using Explain Change Requirements. However, usually any sign of variation in depth of approach must be analysed using more than one basic pattern.

Efficiency of Document Production

Many documents or diagrams produced for the projects are required to have a set structure. While this can be presented in templates, sometimes it is useful for students to have their work analysed according to these rules automatically, rather than repeat the same errors in both their mid-session and final reports. However these rules and also templates change with the domain so must be set up by individual instructors.

Students also recommended that options be set up for them to view each document by its structure alone. That is, the user can select to view the headings and summary only of each section, or headings and descriptions of the role of that section. A template, which includes headings and a description of the role of that part of the document is included as the document template for the course. Students can then edit this directly, with the extra descriptive information removable for displaying and printing a normal view.

The areas where students needed support are:

* Template. To provide document description a template can be included with the suggested sections and a description of the role of each section. The information is formatted in an XML style language that enables the document editor to hide or display information as selected.

* Document rule checker. To improve document design the course documents provide certain expectations for the layout of the document in the form of a template, which can also be encoded as automatic document advice agents. If the students are to be marked on these aspects, they should have them reinforced during their learning.

* View make and use. To improve consistency, the students can link their document by threads of subjects or concepts, which they can display in separate user-selected views.

* Changes. To improve document construction when students are continually changing the document, either when offline or using other software, a summary of these changes can be displayed for group comment. This ensures that other members of the group are made aware of the main changes between versions. A versioning difference list can be generated when in group session, to provide information to generate queries for the students about alterations and present these as part of group discussion.

* Diagram rule checker. This study used Data Flow Diagrams (DFD) in the document, as an example of a diagrammatic form used in the students' projects. However in each discipline different diagrammatic formats are used to represent various parts of planning or design. The rules of each diagrammatic form can be linked to the system through agents, rather than requiring that they be programmed into the tool each time a new diagrammatic format is selected.

Efficiency of Learning

Improving learning also involves linking users to resources that help them change their view of their own understanding of the domain. Diagrams is an area of design that requires feedback in the form of alternative suggestions, since it is difficult for the computer to analyse if a design is good or bad, beyond whether it fits the rules of that format of diagram. In terms of diagrammatic rules, the feedback is uniform for all diagrams so it can be easily implemented by rule-based agents, as done for Entity Relationship diagrams in the software COLER (Constantino-Gonzalez, & Suthers, 2000). With other aspects of the design, such as the taking of alternative design approaches, the feedback will depend on the design that students have developed, so this feedback is difficult to automate.

The three generic areas that patterns are found can be defined as:

* Design. The main request of users is for examples from previous projects. The course projects change each session so the requirement is to find the similarity between each current project and linking them to similar products from previous session projects. Not only does the software have to find resources that are similar to the design being developed by this group, but also has to find this similarity in a different project context. These alternative designs are presented in a way that encourages the user to consider why their design/document differs, and how it is dependent on the context they have assumed. This is similar to other work in distance learning where web courses use hyperlinks to alternative approaches to a problem.

* Interaction. During the entire session online, the users are interacting through the various tools. Often they make poor use of the tools to seek and gain answers to questions, or discuss differences. In particular users are often inexperienced in the steps required to resolve conflict or to even acknowledge and use conflict constructively. Some basic analysis can be made of their use of speech tokens to describe their intention in contributions (7) to the discussion, plus their actions in other tools.

* Learning depth. Similarly the users are often inexperienced in learning course material of any significant depth. Students have been encouraged to learn for assessment and avoid the extra work required to extract meaning from their courses. The main aim and design motivation of the workshop courses, and Intertac, is to motivate and encourage students into a deeper approach and conception of their learning.


This article is about the design of Implementation Patterns. It must be noted that these are different to the interface patterns developed in HCI (Schummer, 2002) as this article looks at the HCH interactions, and are different from the interaction pattern studies in other CSCW systems (McManus & Aiken, 1999), in that the next step of feedback implementation is included in the pattern structure. Also the patterns of approach to learning and the interaction patterns are now combined into one pattern system. These patterns are combined as the conversation or HCH interaction is as significant in the learning process (Pask, 1975) as the concept generation patterns.

Implementation Structure

Since Activity Theory is the analysis used to derive the patterns, an ontology based on this approach is used to describe the pattern structure, using the aspects that came out of the analysis. The aspects of Activity Theory that are important in the analysis are shown in Table 1.

Each pattern is initially assigned a weight, that will be altered by the User Model Agents to be developed in the next version, Intertac-III. The weight determines the agent that acts when there may be more than one agent that has achieved its condition to act.

The aim of developing patterns for learning content is to provide learning objectives in a format that can be easily translated into agent rules. The patterns developed in this analysis are just an extension of the diagrammatic and text structure rules. The structure for the patterns is shown in Table 2.

At present these patterns have been manually extracted and then processed by agents, rather than any automated analysis and learning of patterns from the student actions. This is the limitation of any such rule-based system.

As with any pattern development, it is desirable to analyse the interconnection between patterns. This connection is made through the activity type, in that only one activity should be implementing feedback at any one time. However the role of the patterns in the overall language can be used to select combinations or sequence of feedback. Where two patterns are complementary, they can support each other, where they are contradictory, they negate each other so should not be taking action together.

Pattern Language Structure

The nature of the patterns and the aspects for which they are patterns, form the structure of the pattern language. A pattern language is the semantics of how the patterns related to each other, or are distinct. Alexander, Ishikawa, and Silverstein (1977) used language in reference to architectural patterns in two senses. First, they talked about the shared design language. In the case of an implementation language, this is not always shared between domains. However, this research looks at a range of domains including workflow, document construction, and discussion interactions, providing a broad sample of patterns. The second sense is in terms of an organising principle which facilitates the use of a language. Clearly in implementation of the agents in this work, this structure is important for coordinating a multi-agent system.

The first division in activity type for the patterns is between those that deal with group interactions and those that deal with learning (Figure 2). These patterns can deal with individual or group processes. (8) Most Learning Patterns and all Interaction Patterns fall into the latter category, so unless specified otherwise, patterns are assumed to be for the group process. When users are working alone group and individual become synonymous.


The next division is between agents that scaffold the manipulation of knowledge, provide feedback or encourage the development of learning skills or meta-cognition, such as reflection. Manipulation Agents are again divided into agents that generate actions that can be classified as scaffolding for manipulating conceptual knowledge by arrangement, transformation, deducing, or inducing structure. Meta-cognition agents are divided into agents that generate procedural skills or actions to encourage reflection, or encourage other group processes.


Intertac-II is an extension of the Intertac-I software, using the component-based design to insert agents that implement rules for each application, or as a link between applications, such as taking an overview of a user's contribution. The design of the Intertac-II agents already requires a rule language and structure that can enable the coding of rules from many domains (for instance team work, document formatting, diagram design and requirements engineering). The patterns provide an outline for translating learning needs in a course into an agent process. The next step is to develop a generic agent for each pattern.

The article now looks at the most significant patterns in detail. First, the Design Patterns, which are the most difficult to extract, then the Interaction and Learning Patterns, which are the most generic patterns.

Design Patterns

Design patterns are used to search files from projects in previous years for specific features to be used as alternative examples for students. In this research, the diagrammatic analysis relies on selecting the basic aspects of each drawing primitive and looking for patterns of similar designs. Since the projects differ between years, the design features that are similar are more likely to be structural issues rather than the entire design. The Design Pattern can be used both to check for similar design aspects in previous years and also for differences between an older version of the student group's design and the most recent version. To search files for specific features the agent needs to analyse aspects such as:

* Visual issues, such as joining data too close to each other, is a problem. Alternative better spaced designs can be displayed.

* Keyword searches for keywords missing from the DFD processes and data flows can be done. Unfortunately, the keywords can change between projects in each workshop, so it will be hard to analyse across years, except when dealing with common system processes, such as "login" processes or the designing of time into the system.

* Design steps; some steps in the process of diagrammatic designs are handled badly by students, such as combining sections of a highly detailed level and then moving the detail to a lower level of the design. The two diagrams in this process, before and after, can be displayed from the context of another project, hence abstracted from the details of the particular processes of the student's design.

Similarly, documents from previous years can be searched for changes in patterns. The searched documents also will be from a different software project than the present one. The patterns observed between distinct projects which would be worth noting for search categories are:

* Length of section; where user's sections appear too short on some aspect, display a longer example.

* Use of sections; if important sections such as Nonfunctional Requirements are missed, then the students could be shown an example to see what this section would cover.

* Expert sections; where sections are generally handled badly in the course, such as sections on integration of aspects of the design document, then an example of a skilled report could be displayed.

The Design Pattern uses a unification algorithm on the graphical or textual representation of the data in historical files compared to the present files. Files are searched on the basis of similarity in any one of the previously mentioned criteria, looking for a list of alternatives designs in any second criteria. The alternatives can then be displayed for the student group. (9) The Design Pattern is used to enable the following search types:

* Alternative Design Pattern -- ADP; search for alternatives to the present users' design.

* Change Design Pattern -- CDP; search for changes in design between versions of a design.

* Context Design Pattern -- TDP; search for related aspects of one design throughout the document by keyword search and trace this aspect.

The Design Pattern is described by the structure shown in Table 3.

Interaction Patterns

These are the patterns, which analyse how people interact in the discussion using extra data from students' editing strategies on other tools. The actions that are analysed for patterns are additions, deletions and moves in editing, and the interchange of Tokens in the discussion. The actions are analysed for:

1. time between actions;

2. number of actions of any type;

3. length of the action, such as the edit or the discussion contribution;

4. the user who takes the action and if this changes; and

5. concurrent use of keywords from lexicon.

These patterns are similar to the many previous examples of interaction analysis in Computer Support for Collaborative Learning (CSCL) systems, but are included here to provide a complete pattern language for Collaborative Systems, and enable data from other tools to be included in the interaction analysis. Also, despite the sparcity of information available from text-based dialogue in this simple system, a wide range of interactions can be supported with this added data from actions in other tools.

Learning Patterns

Learning Patterns are those that relate to the depth of learning of a concept or an approach to learning. At present the course timetable enables agents to select the stage the group has achieved to assess the learning depth, or to assess the knowledge that is available to the users to date. Hence this is purely an assessment of the knowledge that is possibly available for synthesis into the users' own understanding.

However this is a very simplistic approach and this research involved a more comprehensive analysis of patterns of the depth of learning. In accordance with the findings of Booth (1992) this research found that:

* Depth of conception is usually attained through exposure to a greater variety of uses of the concept; and

* Depth of approach usually involves the ability to develop an overview of the learning, which is combined with more detailed knowledge.

Hence the Implementation Patterns provide increasing complex representations or experiences of conceptions (Concept Extension Patterns) and ones that monitor bottom-up designs (Complexity Patterns in DFDs) or encourage top-down designs (View Patterns on Documents).


The next stage in the agent development process is to verify the validity of the patterns that are implemented through their feedback. This verification will consider the aspects of Constructive Learning Environments that were adopted as the goals of this work:

* Scaffolding; by developing User (or Group) Models that enable the tracking of students configuration selection (such as role) and overt response to advice and the actions that follow any advice (Kutay & Ho, 2003) some analysis of the scaffolding effect can be examined. Students may respond to such agent advice and it is important that the agents rules are reapplied soon afterwards to verify the effect of the feedback, if any, on the patterns observed.

* Alternatives; another important aspect to verify is the search agents. This will involve running the agents on documents produced by students to verify that the Design Patterns extracted in searches are valid comparisons or alternatives.

* Feedback; a study should be made of the feedback categories that are received during the course of a workshop and how these relate to the resultant document and design produced by the group. This will verify if design problems are missed in the feedback or feedback is made that is not helpful.

* Reflection; during the workshops the students can be interviewed about their approach to learning software design, their approach to working in groups remotely, and their conceptions of the key aspects of the course. These can be related back to the agents that are designed to deal with these learning patterns and verify that the agents have either identified or responded in some way to these approaches.


Constructivist learning environments encourage flexibility and discourage attempts to prescribe actions between students. However by a judicious choice of formats, students can be encouraged to question and expand their understanding from interventions by simple intelligent agents. In particular any learning domain involves either work patterns, rules of design or simply communication patterns, which can be extracted from the data logged by a CSCL system. These patterns and rules can be developed into Implementation Patterns which provide the basis for coding agents to support their use in learning.
Table 1 Activity Theory Structure for Implementation Patterns

Activity or Task Activity enacted in the learning
Rules Rules of analysis of the knowledge concepts or
 approach skills
Information Source Source for the individual knowledge or group
 grounded knowledge which is either the knowledge
 base (log), historical analysed data and
 configuration data (user model) or rules (agents
Temporal Temporal extension of the activity (Akhras & Self,

Table 2 Implementation Pattern Structure

Name To provide easy reference.
Activity Type See Figure 2.
Information Source of data User or Group, and the Tool this pattern is
 related to.
Rules Outline Problem to be solved or skill/concept
 learned by feedback.
Focus The learning aspects or interaction aspects
 that are the focus of this pattern.
Information Source of What data interplay signifies this pattern
 Conditional is achieved.
Action Solutions to problem and processes to follow
 for feedback. May involve a series of steps,
 if the first does not get desired action by
 group or user, try next in order.
Information Source of Goal Desired end result or response for each
Example Practical and specific.
Weight Initially assigned on basis of pattern
Semantics Role this feedback plays in the pattern
 language semantics.

Table 3 Design Pattern Structure

Name Design Pattern.
Activity Type Alternative/Changed/Traced Document/Diagram.
Features Search specifications.
Weight Significance of this difference.

Funding Source

This research was generously funded by the Faculty of Engineering, The University of New South Wales, as an Educational Research Project.


The authors would like to thank: David Abdelmassih, Manoj Chandra, Sherman Lo, Zhicong Leo Liang, Zi Qi Lu and Naveed Hussain for contributing ideas and their programming skills.


(1) See Booth (1992) for similar research into learning needs for students learning programming.

(2) Similar work has been done in single-user Intelligent Learning Environment by Akhras and Self (1997).

(3) See Barros, Verdejo, Read, & Mizoguchi (2002) for extended learning ontology.

(4) Where an Implementation Pattern is described, it is written with initial capitals.

(5) Similarly, any analysis of student progress in learning by the agents can be edited to improve student learning (Kay, Halin, Ottomann, & Razak, 1997).

(6) The projects involve developing software for a realistic client, based on the requirements the client has provided.

(7) Compare to the work of McManus and Aiken (1995).

(8) Individual interaction with the group is treated as an Individual process.

(9) Rule Checkers should be applied also before the alternative designs are displayed.


Akhras, F.N., & Self, J.A. (1997). Modelling learning as a process. Proceedings of the 8th World Conference on Artificial Intelligence in Education (AI-ED'97), Kobe, Japan.

Akhras, F.N., & Self, J.A., (2000). System intelligence in constructivist learning. International Journal of Artificial Intelligence in Education, 11, 344-376.

Alexander, C., Ishikawa, S., & Silverstein, M. (1977). A Pattern Language, Oxford: Oxford University Press.

Barros, B., Verdejo, M. F., Read, T., & Mizoguchi, R. (2002). Applications of a collaborative learning ontology. Proceedings of MICAI'2002 Mexican International Conference on Artificial Intelligence, LNAI, (pp. 301-310). New York: Springer-Verlag.

Bayle, E., Bellamy, R., Casady, G., Erickson, T., Finicher, S., Grinter, B., Gross, B., Lehder, D., Marmolin, H., Moore, B., Potts, C., Skousen, G., & Thomas, J. (1998). Putting it all together: Towards a pattern language for interaction design. A CHI'97 Workshop. SIGCHI Bulletin, 30(1) 17-23.

Booth, S. (1992). Learning to program: A phenomenographical perspective. Goteborg Studies in Educational Sciences, '89, Acta Universitatis Gothoburgensis.

Constantino-Gonzalez, M., & Suthers, D.D. (2000). A coached collaborative learning environment for entity-relationship modelling. In G. Gauthier, C. Frasson, & K. Van Lehn, (Eds.), Intelligent Tutoring Systems, Proceedings of the 5th International Conference (ITS 2000), (pp. 325-333). Berlin: Springer-Verlag.

Constantino-Gonzalez, M., & Suthers, D.D. (2001). Coaching collaboration by comparing solutions and tracking participation. In P. Dillenbourg, A. Eurlings, & K. Hakkarainen (Eds.), European Perspectives on Computer-Supported Collaborative Learning, Proceedings of the First European Conference on Computer-Supported Collaborative Learning, (pp. 173-180). Universiteit Maastricht, Maastricht, the Netherlands.

Constantino-Gonzalez, M., Suthers, D., & Escamilla de Ios Santos, J.G. (2003). Coaching web-based collaborative learning based on problem solution differences and participation. International Journal of Artificial Intelligence in Education, 13(2-4), 263-299.

Dillenbourg, P., Baker, M., Blaye, A., & O'Malley, C. (1996). The evolution of research on collaborative learning. In E. Spadia & P. Reman (Eds.), Learning in human and machine: Towards an interdisciplinary learning science, (pp. 189-211). Oxford, UK: Elsevier.

Kay, J., Halin, Z., Ottomann, T., & Razak, Z. (1997). Learner know thyself: Student models to give learner control and responsibility. Proceedings of the International Conference on Computers in Education (ICCE'97) (pp. 17-24), Malaysia. Charlottesville, VA: Association for the Advancement of Computing in Education.

Kutay, C. (2003a). Intertac. Software architecture, technical Report No. 0318. Computer Science and Engineering, University of New South Wales. Retrieved from, 2003

Kutay, C. (2003b). Implementation patterns for supporting learning and group interactions. Unpublished doctoral dissertation, The University of New South Wales, Sydney.

Kutay, C., & Ho, P. (2003). Assisting students by analysing their interactions and learning, submitted to AIED'03 Workshop on Learner Modelling for Reflection, Sydney University.

McManus, M.M., & Aiken, R.M. (1995). Monitoring computer-based collaborative problem solving. International Journal of Artificial Intelligence in Education, 6(4), 307-336.

McManus, M.M., & Aiken, R.M. (1999). Petri nets: Using a formal model to describe synchronous rules in concurrent processes. Technical Report TUCIS-TR-1999-001, Temple University.

Maiden, N.A.M., Cisse, M., Perez, H. & Manuel, D., (1998, June). CREWS validation frames: Patterns for validating systems requirements. Fourth International Workshop on Requirements Engineering: Foundation for Software Quality (RESFQ), Pisa, Italy.

Martin, D., Rodden, T., Rouncefield, M., Sommerville, I., & Viller, S. (2001). Finding patterns in the fieldwork. Proceedings of ECSCW 2001, (pp. 39-58), Bonn, Germany.

Muhlenbrock, M., Tewissen, F., & Hoppe, H.U. (1998). A framework system for intelligent support in open distributed environments. International Journal of Artificial Intelligence in Education, 9, 256-274.

Nardi, B. A. (Ed.) (1996). Context and consciousness. Activity theory and human-computer interactions. Cambridge, MA: MIT Press.

Papatheodorou, C., Vassiliou, A., & Simon, B. (2002). Discovery of ontologies for learning resources using word-based clustering. Proceedings of the World Conference on Educational Multimedia, Hypermedia & Telecommunications (ED-MEDIA 2002), (pp. 1523-1528), Denver.

Pask, G. (1975). Conversation, cognition, and learning. New York: Elsevier.

Savery, J.R., & Duffy, T.M. (1995). Problem based learning: An instructional model and its constructivist framework. Educational Technology, 35(5), 31-7.

Schummer, T. (2002). PATORG. Retrieved from

Simone, C. (1994). Supporting collaborative dialogues in distance learning. In M. F. Verdejo & S.A. Cerri (Eds.), Collaborative dialogue technologies in distance learning. NATO ASI Series, F133 (pp. 156-169). New York: Springer-Verlag.

Wood, D. (2001). Scaffolding, contingent tutoring and computer supported learning. International Journal of Artificial Intelligence in Education, 12, 280-292.


University of New South Wales, Sydney, Australia
COPYRIGHT 2005 Association for the Advancement of Computing in Education (AACE)
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2005, Gale Group. All rights reserved. Gale Group is a Thomson Corporation Company.

Article Details
Printer friendly Cite/link Email Feedback
Author:Ho, Peter
Publication:International Journal on E-Learning
Geographic Code:8AUST
Date:Jan 1, 2005
Previous Article:Supporting learning and information sharing in natural resource management with technologies for electronic documents.
Next Article:A virtual hyperbooks model to support collaborative learning.

Related Articles
Promoting interactive learning through contextual interfaces on a Web-based guided discovery CAL.
Permutations of control: Cognitive considerations for agent-based learning environments.
SIVOX introduces RealCall. (New Products).
Envision Telephony integrates products into performance suite.
The effects of different levels of interaction on the achievement and motivational perceptions of college students in a web-based learning...
Designing a web enhanced multimedia learning environment (WEMLE) for project management.
Effect of a Socratic animated agent on student performance in a computer-simulated disassembly process.
Reliability and factor structure of the Attitude Toward Tutoring Agent Scale (ATTAS).
Multimedia integration in online courses.
Online learners take the lead in self-direction.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |