Multiagent Framework for Course Personalization

---

Context

The design basis for this framework is a university e-learning environment (i.e., Athabasca University), where learners are taking courses in an asynchronous mode (grouped or individual). In this context, most learners are working and have a personal need to enhance their careers. As well, these learners often have a long-term learning plan and prefer a flexible and individualized learning environment. A significant percentage of these learners is always mobile and needs to have access to their courses from everywhere, anytime, and by using different devices. Thus, the developers of this project must consider the following factors:

• Asynchronous e-learning

• Mobile users

• Multidevices environment

• Adapted courses

• Personalized interfaces that offer the same look and feel

System Architecture

In this context, we are proposing a multiagent architecture for implementing an e-learning system that offers course personalization and supports mobile users connecting from different devices. The detailed architecture of the system is articulated around five main components:

1. User profile repository: For each user, the system maintains a profile that has two components: the learner’s model, and the user’s preferences regarding learning style, interfaces, and content display.

2. Device profile repository: For each device, the system maintains a profile of the features and capabilities useful for providing the e-learning service (screen size, bandwidth limit, colors, resolution, etc.). Some features that can be automatically detected by the system (operating system, browser, plug-ins) are not stored in the repository but are integrated with the profile when initializing the terminal agent.

3. Learning object repository: This contains the course’s teaching material defined as learning objects (Wiley, 2000).

4. Course database: For each course, the system maintains two knowledge structures: the course study guide and the course study plan.

5. Multiagent system composed of stationary and mobile agents.

Presented in Figure 1 are the main components of the system.

Figure 1: System architecture

Profile Manager Agent

The profile manager is implemented as a stationary agent. It manages the knowledge related to the learners and all defined devices (terminals). The main tasks of the profile manager are as follows:

• Performing user authentication.

• Acting as a central register, where each new learner must be registered.

• Managing and assuring the consistency of the databases containing the learner and device profiles.

• Receiving service requests from terminals and giving access to the user profile data.

• Initiating and sending the user agent and terminal agent to the remote device.

• Checking the versions of the user agent and the terminal agent that reside on remote terminals and automatically downloading any necessary updates.

Course Provider Agent

This is a stationary agent that manages the knowledge about courses and teaching strategies. The main tasks of the course provider are as follows:

• Providing an interface for defining learning objects and course knowledge (study guide and study plan).

• Receiving service requests from terminals and giving access to course data.

• Generating the course study guide and study plan based on the user profile and the teaching strategy.

• Packaging the course teaching material according to the user profile and device profile.

• Initiating and sending the tutor agent and terminal agent to the remote device.

User Agent

This is a mobile agent that carries and manages a local copy of the user profile (user’s preferences and learner’s model) to the remote terminal. The main tasks of the course provider are as follows:

• Providing the tutor agent and terminal agent with the user information (profile, identification).

• Managing and synchronizing user profile duplication with the central server.

• Providing the local personalization of the course material—In collaboration with the terminal agent and the tutor agents, the user agent insures the display of the course material according to the user’s preferences and the terminal’s capabilities.

Tutor Agent

This is a mobile agent that manages the course delivery to the roaming user. The main tasks of the course provider are as follows:

• Carrying and managing the course material and study plan.

• Providing a personalized learning service to the learner based on his or her model and learning style.

• Insuring the adaptation and packaging of external course content— Because the system is open to third-party providers, external course material will need to be converted to the required format and adapted to the user and device profiles. In collaboration with the user agent and the terminal agent, the tutor agent insures the necessary adaptation and conversion.

• Synchronizing the course content with the server.

Terminal Agent

This is a mobile agent that maintains the terminal profile of the corresponding device and insures the display of services according to the user’s preferences and terminal’s capabilities.

Knowledge Structures for Adaptive Course Delivery

Course’s Study Guide

The course content is organized around a set of concepts. Each concept has teaching material associated with it that comes from either an external source or a learning object. The course study guide defines the relationships among these concepts. The relationships consist of prerequisite, similar, and substitute relationships.

Course’s Study Plan

The course material is organized in terms of units and sections. In each section, a set of concepts is learned by using different tasks: readings, labs, and tests. The study plan defines the sequencing of the course content and the time constraints and deadlines for the different tasks of the learning process.

Learner’s Model

This is a fuzzy overlay model (Kurtz et al., 1990) based on the course concepts. It represents static beliefs about the learner and, in some cases, is able to simulate the learner’s reasoning. Each concept in the model is associated a fuzzy value representing the assessment of the learner’s knowledge regarding this concept.

The system uses two different versions of the learner’s model: a global model and a local model. The global learner’s model is stored within the user profile repository and represents concepts reported to the system about the learner or learned from the learner’s past experience with different courses. As well as representing the concepts and associated fuzzy values, this model represents the relationships among concepts (prerequisite, similar, and substitute). The local learner’s model is managed by the user agent within the user’s terminal and is related to a specific course. This model represents only the course concepts and associated values.

The local learner’s model is refined based on the learner’s interaction with the system when reading the course material and doing the assessment exercises.

The local model is also used to update the global learner’s model and is initialized from the global model.

Teaching Strategy

This is a set of rules that controls the adaptation of the course. The teaching strategy consists of rules for sequencing the course-material components, adding or dropping course-material components, and selecting between similar or equivalent course-material components.

Adaptation of Third-Party Provider Content

Unlike traditional systems, our system will be open to third-party providers. In fact, we aim in the future to implement an infrastructure (e-market place) to provide collaborative e-learning services. Thus, we need to implement a process that provides user-side device independence for Web content. The main idea behind this process is to construct a basic generic page from the source and then to mark up that document with appropriate tags, as determined by the user profile and the device profile.

A Web course’s content always involves different resources (files, database, learning objects, etc). Therefore, the adaptation process consists of creating a Java Servlet or JSP document that connects to data sources and objects and produces an XML document. The main idea here is to use a two-stage process that generates in the first step an XML document (model), and then translates the generated model to a rendering format (HTML, WML, etc.) that will be presented to the user.

Described in Figure 2 are the two stages of the services.

Figure 2: Adaptation of external course content

The first step in the content module is to create an XML document from the content resources. If we use the logic of Web services, this activity will correspond to the model creation service (Figure 2), which implements an XML generator by using JSP or Java Servlet, which generates an XML document of the Web content. The user profile is used here to personalize the content according to this profile. The second step in the content module is to create a rendering format for the XML document. This activity corresponds to the view transformation service (Figure 2), which implements an XML transformer by using XSLT or DSSSL, which generates a rendering format for the XML document. Because the rendering format depends on the devices’ features and the user’s preferences, the user profile and device profile will be used in this process.

The two stages of the services will provide us with more flexibility and device independence than would be possible otherwise:

• The separation of the service model from the service view will provide us with device independence and facilitate the maintenance of the content- generation process.

• With browsers including a W3C-compliant XSLT engine, more processing will occur on the client side and reduce the work done by the server.

• The services may be distributed over several machines, if needed, to balance the overall load.

---