Jean Girard, Gilbert Paquette, Alexis Miara, Karen Lundgren
LICEF Research Centre, Télé-université
1001 Sherbrooke St. East, Montreal
Abstract. Intelligent assistance in telelearning environments is even more important than in individual tutoring systems because of the inherent complexity of distance education. But the problem here is quite different and provides large areas of unexplored territories especially in the full exploitation of the multiple data sources captured from the interaction of the different actors involved in a telelearning event. We will address some of these questions by presenting first a model of a Virtual Learning Centre (VLC) and an implementation for Web-based training called Explora. A VLC focuses on the interaction spaces between five theoretical actors: the learner, the informer (content expert), the trainer, the manager and the designer. We will give an example of such an environment and show how the VLC supports learners and the other actors in such a case. Then we will focus on a three-dimensional assistance space in a VLC based on a typology of assistance resources. Finally, methods and tools to build an intelligent advisor for a web-based Telelearning environment will be discussed using an operational JAVA implementation on the Internet.
Key words. Learning environments and micro-worlds, Non-standard and innovative interfaces, Student modelling
1. The Case for Advisor Agents in a Virtual Learning Centre
We live in societies coping with an exponential growth of information. In the knowledge society, new competencies and higher level skills are required. The rapidly evolving availability of multimedia telecommunication is a challenging answer to this knowledge acquisition and knowledge building gap. But we have to integrate many types of resources to really enhance learning. We see a telelearning system as a society of agents, to use Marvin Minski’s term, some of them providing information, others constructing new information, stills others helping collaboration between agents or providing assistance to the other agents on content, pedagogical process or organisation of activities.
What is behind terms like “distance education”, “on-line learning”, “telelearning”, “multimedia training” is a multi-facetted reality from which we can identify six main paradigms: the open classroom integrating technologies in traditional classrooms, the virtual classroom [1,2], the teaching media, focused on multimedia courses on a CD ROM, Web-based training 3, On-line learning communities [4,5 and Electronic performance support system (EPSS) .
Intelligent assistance, based on knowledge of the learner’s activities and production, is even more important in telelearning systems because of their complexity. But the problem is quite different than in individualised ITS research. It provides large areas of unexplored territories especially in the full exploitation of the multiple data sources captured from the interaction of the different actors involved in a telelearning event.
2. EXPLORA, a Web-Based Virtual Learning Centre
Our Virtual Learning Centre model 10 emphasises the concept of process-based learning scenario coupled with assistance resources. Basically, the learner proceeds into a scenario, a network of learning activities, using different kinds of information resources to help her achieve the tasks and produce some outcome: a problem solution or new information that can be used in other activities. The assistance resources for each task are also planned at design time. The assistance can be distributed among many agents: trainers interacting through email or teleconferencing, other learners, contextual help or intelligent advisors.
2.1 Actors, roles and agents
We have described elsewhere [7,8,9] how we have built an object oriented model of a Virtual Campus using software engineering methodology. In our Virtual Learning Centre architecture, we identify five actors.
The Learner transforms information into personal knowledge.“ Information ” here is any data, concrete or abstract, perceptible by the senses and susceptible of being transformed into knowledge. “ Knowledge ” means the information that has been integrated by a cognitive entity into its own cognitive system, in a situated context and use.
The Informer (the content expert) makes the information available to the learner. It may be a person or a group of persons presenting information to the learners, but also a book, a video, a software or any other material or media.
The Designer is the actor building, adapting and sustaining a learning system (LS) that integrates information sources (human informers or learning materials), and also self-management, assistance and collaboration tools for the other actors.
The Trainer facilitates learning by giving advice to the learner about his individual process and the interactions that may be useful to him based on the learning scenarios defined by the designer.
Finally, the Manager facilitates learning by managing actors and events, for example creating groups or making teleservices available in order to insure the success of the process, based on the scenarios defined by the designer.
Figure 1 - Actors and interaction spaces
2.2 Interactions spaces and resources
Figure 1 shows the five theoretical actors and their interactions. We will limit ourselves to interactions in which the learner is involved while learning, at delivery time.
Interactions between learner and designer. These are the interactions where the learner interacts with the learning environment into which the designer has in a way “mediated” himself by creating it. Here, the learner is preoccupied with the self-management of the learning activities, of their input resources and of the products he has to build.
Interactions between learner and informer. These are the interactions where the learner, individually, consults the information made available by the informer, and process them in the production space to produce certain results while building personal knowledge.
Interactions between learners. These are interactions using different forms of collaboration or cooperation between learners for team work, group discussion, collaborative problem solving, etc.
Interactions between learner and facilitators. These interactions concern the assistance that the system can provide to the learner on both the pedagogic (from the trainer) and the management (from the manager) dimensions of telelearning. We will study these interactions in the next section.
2.3 Host systems and HyperGuides
At delivery time, the learner and the other actors interact within a computerised learning environment. This host system presents the content of a learning event, proposes activities, identifies resources to achieve them and describes productions to make.
Figure 2 – A host telelearning environment and a trainer’s HyperGuide within Explora
Figure 2 present such a web-based host system that intends to help learners build knowledge about the job market, their job objectives, and methods to apply while searching for a job.
Actors need different point of view on the host system in each interaction space: self-management, information-production, collaboration or assistance. For example, in the information space, a learner will need different input resources such as a list of related web sites, descriptions of job categories, questionnaires to identify his skills or qualities, etc. In the same information space, a trainer needs other information resources: traces of the learners activities for diagnosis, information on the group of learners, information on learner productions, annotation tools to identify and organise information for assistance.
These resources are made available through an external palette as shown in the floating window on the right of figure 2. This is what we call an HyperGuide. It is an actor’s environment for a course or program supported by the Virtual learning centre. It groups resources into five interaction spaces (self-management, information, production, collaboration and assistance) according to an actors’ role and course specificity. The resources can be generic tools developed specifically for the Virtual Learning Centre, they can be generic commercial tools or they can be web resources (used in many courses) such as a technical FAQ or a virtual library.
The following table shows a possible distribution of resources into the production and the assistance spaces, for the Job search example, for the learner, the trainer and the designer.
ISA Design Workbench
Bank of design objects
WebCT Authoring Tool
Learning System Editor
Advisor Definition Tools
Access to trainer/manager
Access to content experts
Access to course designers
FAQ on tutoring
Instructional design advisor
FAQ on ISD methodology
Help resource persons
Table 1 – Example of the distribution of resources in the Virtual learning centre
These resources are external to the web course but some of them, for example the learner’s trace, content navigator or intelligent advisors are linked to a course by the designer using the learning system editor and the advisor definition tools.
There are many advantages to such an architecture. The Virtual Learning Centre is at the learning organisation’s level, thus avoiding duplication and facilitating evolution and reuse of resources from one course to another. It also speeds up the design process because each individual web-course if freed from all the generic resources and the circulation of information management between different actors.
3. Assistance in a Virtual Learning Centre
Whatever the agents, human or computer-based, the assistance must be « intelligent », that is, informed of the user, of the kind of tasks he is involved in, of the information he has consulted or produced, of the interactions and collaboration with others, and finally of his use of the assistance resources. In another words, the central question of ITS research, the user model, reappears in telelearning systems, but in a different way.
We will first to present a typology of possible assistance resources, including advisor agents. Then we proceed with the preliminary step of course modelling and integration between the HyperGuide and the Web course. We finally present the way the user model is constructed and updated at delivery time and give some examples of the pieces of advice for a particular telelearning environment.
3.1 Types of assistance resources
When the designer plans the telelearning system, he must select or build different kinds of resources for the assistance space. Assistance resources can be addressed to different actors, for different purposes and by different means, yielding a three dimensional assistance space.
The preceding discussion gives a framework for these decisions.
The theoretical actors to which assistance is addressed is the first dimension of the typology. Often the organisation will distribute roles differently. For example, the trainer and the informer roles car be merged; as a consequence, there might be only four assistance spaces to design. On the other hand, the designer’s roles can be split into more specialised roles such as content modelling, pedagogical design and learning material production, each of these actors having different assistance needs.
The second dimension of assistance, its object, can be related to each of the interaction spaces: self-management, information, production, collaboration or instructional and organisational assistance. The assistance can be given to help users progress through the activities, acquire consistent information and knowledge, engage in peer collaboration or use efficiently the available assistance resources.
Finally, the third dimension identify in what way the assistance will be provided: by a human facilitator, by an access to help desks, using FAQs and help files, providing contextual on-line help or intelligent advisor agents.
Because of its complexity, to design assistance in a telelearning system is a huge task. To reduce this complexity, we propose to use a design method that help identify the appropriate resources based on a good understanding of their possibilities and limitations. Also, the integration of multiple assistance resources and agents, especially human facilitators reduces the load on heavily computerised resources such as an ITS. Finally, the development of generic assistance resources at the VLC level with the help of advisor building tools should help tackle the heaviest components of the assistance space.
3.2 Modelling a course
With these kinds of problems in mind, we have developed a method for learning systems engineering called MISA 11. The MISA method presents the ID processes and tasks according to an engineering perspective analogous to software engineering.
The method is a complex process decomposed at several levels, into sub-processes. Each sub-process has its inputs and its 30 main products and 150 sub-tasks well defined, the whole process generating a learning system as its final output. This method innovates by using cognitive modelling techniques to represent knowledge, as well as pedagogical models, learning materials model and delivery models. These four aspects of a learning system, are clearly differentiated but they also are interrelated through specific associations in each of six main phases, making the engineering process visible and structured, thus facilitating quality control of the processes and their products.
For the assistance space, we need to focus here on the design of the knowledge model representing the content to be learned and the instructional model representing the learning events (program, course or activities). The terminal learning events are called learning units for which we design an instructional scenario corresponding to a target population of learners. These learning events form a network into which the learner will navigate. Similarly, the knowledge model is a network of facts, concepts, procedures and principles that the learner must acquire/build.
Actually, in the implementation presented here, we have not yet considered the full potential of such MOT models for advising. We simplify the problem par reducing both models to hierarchical trees. The first one is the instructional structure (IS) representing a curriculum, its main subdivisions down to terminal learning events, that is learning activities which are the main components of a learning scenario. The second one is the knowledge structure (KS) compose of a main knowledge object, decomposed into related parts and ending with simple objects. These models can be displayed as generic VLC tools. On figure 3, such a tool is presented for the instructional model navigator.
Figure 3 – VLC tools: an instructional model viewer and a collaborative path finder
To integrate a course like the one in figure 2 to the VLC, we need to link each unit in the IS and the KS, and progress level within them, to actual web pages or sections in the host system. This will enable the system to capture the user’s action and build a user model. Then this user model will offer an alternate way for any user to navigate in the course by selecting a learning event in an instructional model viewer (figure 3 - left), asking to display the corresponding pages or sections. Also, another VLC tool (figure 3 – right) enables a user to see another’s progress in the same web course, to help collaboration as well as tutoring.
3.2 The user model
The integration of a course into the VLC is done through a simple interface where the designer of the advisor describes the IS and the KS. Also, another design tool helps define conditional principles that will update the user model, display an advice or engage in a dialogue with the user. Finally a management tool identifies the learners and facilitators assigned to the course, making possible, though JAVA scripting, to navigate into pages visited by a trainee or a co-learner.
With the help of these tools, the designer of the advisor will go through the following steps.
Define the IS as a hierarchical list of instructional units IU1, IU2,....., IUn and the KS as a hierarchical list of knowledge units KU1, KU2,....., KUm
Define for each IU or KU a set of ordered symbols called progress levels Ip1, Ip2,....., Ipk ; Kp1, Kp2,....., KpL and assign user events in the host system to each couple formed of a IU or KU with one of its progress level.
Define how the acquired or desired progress levels will be defined at the beginning and how they will be updated according to the user’s actions.
Define the conditions that will fire each advice or action and state the piece of advice or describe the action using a symbolic language.
At any time the system evaluates the user’s actions in the host system and assigns, for each IU or KU, an estimated progress level considered to be acquired by the user, called his belief, and a targeted progress level called his goal. The user model at time t is simply the set of all beliefs and goals assigned at timet to each IU or KU.
User-Model (t) = BIU(I,t) , GIU(I,t)I=1,n BKU(J,t) , GKU(J,t)J=1,m
where BIU(I,t) and GIU(I,t) are the acquired and desired progress level for IU number I at time t
and BKU(J,t) and GKU(J,t) are the acquired and desired progress level for KU number J at time t
The user model is updated essentially in three ways: by the designer’s predefined rules, by querying the user at run time and by some action the user can take to modify the model.
First, the designer will predefine basic actions on the model, that is principles stating that if certain conditions are met, the model will be updated to some belief or goal level for a certain unit (IU or KU). Actually, the following actions can be taken: add or suppress a progress level in a unit; move up or down the progress level in a unit, query the user with a message and offer a set of possible answers.
A second way to update the model is when the user is queried. Upon certain answers, other questions can be asked, until the end of a predefined decision tree. Then, depending on the path of the user’s answers, the system will be able to update the model according to the designer’s definition in such a case.
Finally, the user can see at any time the different belief and goal levels assigned to him for any IU or KU. The VLC tool shown on figure 4 is one way of doing this. According to some evaluation of the distance between a progress level and the next one, and the number of progress levels for a IU or a KU, different weights can be assigned to progress levels. In this way, the system can display a bar diagram showing the proportion of progress in each IU or KU according to the belief level.
Figure 4 – A progress viewer in the VLC
Such a display can help the learner orient his actions. Also, he can disagree with the system. The tool of figure 4 enables him to change the values of any acquired or goal progress level.
4. Extending the advisory system
We will now conclude this paper by identifying extensions to the actual implementation we have outlined here. There are three directions in which we want to move.
The first one in to extend the actual advisor to support collaboration. Right now, the JAVA implementation enables a user to see the other users’ learning path, look at their progress in the host environment and communicate with them accordingly. An extension of the user model has been designed by the first author by adding another couple of progress level to the model for each instructional unit (IU) or knowledge unit (KU) called social belief and social desire 12. These values identify the believed capability of a user to interact with others at a certain level for a given IU or KU, and also his intention to do so (is it a goal?). Based on this extension of the user model, a syntax has been defined to update this model, making it possible to advice on collaborative issues such as the selection of peer learners, the selection of tasks on which to collaborate, or the identification of knowledge on which to exchange.
The second set of tasks we now face is to design advisors for other actors than the learner, particularly for the trainer and the designer. The actual learner’s advisor is based on the use of VLC viewers that help the learner use his user’s model and act accordingly. Such tools will have to be tailored for the trainer’s role; for example, to help him make accurate diagnosis of the learner or groups of learners, to provide meaningful pieces of advice, to evaluate the learner’s achievements. Also we will work on an advisor for the designer, to help him adapt his scenario according to the characteristics of a design project, to navigate efficiently within the learning systems engineering method (MISA) and to assess the consistency and the quality of a learning system design. We will here extend previous work on the Epitalk architecture 13.
Finally, we need to improve the design tools to help build such advisors. Defining the actions and pieces of advice is a time consuming task. We believe that the approach presented here can help automate and systematise a good part of the job. For example, once the design of the scenarios is done using a method like MISA, the IU and KU hierarchical lists can be produced automatically, default progress levels and updating actions can be proposed to the designer, action and advice frames can also be proposed. When this is done, we would like also to replace the hierarchical lists for the IS and the KS with richer instructional or knowledge models.
The authors wish to underline the contribution of Claude Ricciardi-Rigault, Chantal Paquin, Ileana de la Teja, Fréderic Bergeron and of all the other researchesrs who participate in the various Virtual Campus projects at LICEF and have helped these ideas to mature. Also, a special thank to the Quebec Information Highway Fund, the TeleLearning Network of Centers of Excellence and the Office for Learning Technology (OLT) who have contributed to the funding of these projects.
1 Wilson J. & Mosher D. The Prototype of the Virtual Classroom. Journal of Instructional Delivery Systems, Summer 1994, 28-33.Wilson J. & Mosher D.
2 Hiltz, S. Evaluating the Virtual Classroom, in Harasim, L. (ed.) Online education: perspectives on a new environment. New-York: Praeger Publishers, 133-184, 1990.
3 Hall, B. Web-based training, 1998
4 Harasim, L. Online education: An environment for collaboration and intellectual amplification. in L. Harasim (Ed.), Online education: Perspectives on a new environment. New York: Praeger Publishers, 1990.
5 Ricciardi-Rigault, C., Henri, F. Developing Tools for Optimizing the Collaborative Learning Process, Proceedings of the International Distance Education Conference, June, Penn State, 1994.
6 Gery G. Granting three wishes through Performance-Centered Design NATO Communications of the ACM, volume 40, number 7, pp.54-59, July 1997.
7 Paquette G., Bergeron G., Bourdeau J. The Virtual Classroom revisited, TeleTeaching'93 Proceedings, Trondheim, Norway, août 1993.
8 Paquette G., Modeling the Virtual Campus. In “Innovating Adult Learning with Innovative Technologies” (B. Collis and G. Davies Eds) Elsevier Science B.V., Amsterdam, 1995.
9 Paquette G. , Ricciardi-Rigault C., Paquin C., Liégeois S. and Bleicher E. Developing the Virtual Campus Environment, ED-Media International Conference, Boston, USA, June 1996.
10 Paquette, G. Virtual Learning Centers for XXIst Century Organizations. In F. Verdejo and G.. Davies (eds.), The Virtual Campus (pp. 18-34). Chapman & Hall, London, 1997
11 Paquette G. , Aubin C. and Crevier, F. Design and Implementation of Interactive TeleLearning Scenarios. Proceedings of ICDE’97 (International Council for Distance Education), Penn State University, USA, June 1997.
12 Girard J., Paquette G., Giroux S. Architecture de Système Conseiller multiagent sur la collaboration dans un Système d’apprentissage. LICEF Research Reports, Télé-université, 1997.
13 Paquette G. and Girard J. AGD: a course engineering support system, ITS-96, Montreal, June 1996.