Related Works | Designing Distributed Learning Environments with Intelligent Software Agents

< Day Day Up >

Overview of Some Implemented Systems

Since the early days of Internet expansion, researchers have implemented different kinds of adaptive and intelligent systems for Web-based education. Almost all of these systems inherited their features from the two well-known types: Intelligent Tutoring Systems (Brusilovsky, 1995) and Adaptive Hypermedia Systems (Brusilovsky, 1996).

Intelligent tutoring research focuses on three problems: curriculum sequencing, intelligent analysis of learner’s solutions, and interactive problem-solving support. Adaptive hypermedia systems research focuses on adaptive presentation and adaptive navigation support. In this section, we briefly present some implemented systems that use one or more of these concepts. For more details on these systems, the reader can refer to the cited references.

ELM-ART (Weber & Specht, 1997; Weber & Brusilovsky, 2001): This is an on-site intelligent learning environment that supports example-based programming, intelligent analysis of problem solutions, and advanced testing and debugging facilities. ELM-ART II supports active sequencing by using a combination of an overlay model and an episodic user model. The overlay model represents the student’s problem-solving knowledge and consists of a set of goal-action or goal-plan rules. The episodic model (Weber, 1996) uses a case-based approach and consists of cases describing problems and solutions selected or developed by the student. ELM-ART II also implements adaptive navigation based on the student’s model. Adaptive navigation uses a traffic-light metaphor for visual annotation of links. Green, red, yellow, and orange balls are used to annotate the links to the next pages to be visited. Finally, ELM-ART II supports example-based problem solving. It encourages students to reuse the previously analyzed examples for developing new solutions. ELM-ART II can predict the student’s way of solving a particular problem and find the most relevant examples from his or her individual learning history.
ACE (Specht & Oppermann, 1998; Specht, 2000): ACE is a Webbased intelligent tutoring system that combines instructional planning and adaptive media generation to deliver individualized teaching material. The ACE uses three models for adapting different aspects of the instructional process: domain model, pedagogical model, and learner model. ELMART II was basically the starting point for ACE. Hence, ACE inherited many knowledge structures from ELM-ART II. The learner model of ACE combines a probabilistic overlay model and episodic model similar to those used in ELM-ART II. The probabilistic overlay model is used for several adaptation levels: adaptive sequencing, mastery learning, adaptive testing, and adaptive annotation. The episodic model is used to generate hypotheses about the learner’s knowledge and interests. The domain model describes the domain concepts and their interrelations and dependencies. It is built on a conceptual network of learning units, where each unit can be either sections or concepts. The pedagogical model contains the teaching strategies and diagnostic knowledge. The teaching strategies define the rules for different sequencing of each concept in the learning material. The diagnostic components store the knowledge about several types of tests and how they have to be generated and evaluated. ACE supports adaptive navigation by using adaptive annotation (cf., ELMART II) and incremental linking. It also supports adaptive sequencing, adaptation of unit sequencing, and teaching strategy. Finally, ACE implements a pedagogical agent that can give individualized recommendations to students depending on their knowledge, interests, and media preferences (Schoech et al., 1998).
InterBook (Brusilovsky et al., 1998; Brusilovsky & Eklund, 1998): This is a tool for authoring and delivering adaptive electronic textbooks on the Web. InterBook supports adaptive sequencing of pages, adaptive navigation by using links annotation, and adaptive presentation. Adaptive sequencing and navigation are implemented by using a frames-based presentation that includes a partial and adaptive table of contents, a presentation of the prerequisite knowledge for the current page, and an overview of the concepts the page discusses. Interbook also offers some special features for supporting the learning process. They include a glossary and a “teach-me” guided tour generator for learning any specific concept. For its implementation, InterBook uses the same approach and architecture as ELM-ART II.
DCG (Vassileva, 1997; Vassileva & Deters, 1998; Brusilovsky & Vassileva, 2003): This is an authoring tool for adaptive courses. It generates personalized courses according to the student’s goal and model and dynamically adapts the course content according to the acquired knowledge. The DCG supports adaptive sequencing by using a domain concept structure, which helps in generating a plan of the course. The DCG uses the concept structure as a road map for generating the course plan. A planner is used to build the course plan by searching for subgraphs that connect the concepts known by the learner to the new goal concept. The course sequencing is elaborated by linearizing the subgraphs using the pedagogical model. The pedagogical model contains a representation of the instructional tasks and methods and a set of teaching rules. The DCG uses two instances of the student’s model, one on the server side updated only after closing the learning sessions, and a more dynamic one on the client (learner’s) side. The learner’s model is represented as an overlay with the concepts structure and contains the probabilistic estimations of the student’s level of knowing the different concepts. For students who fail to learn a concept, DCG offers two levels for replanning the course: local plan repair and global replanning. Local plan repair is used to change only part of the plan related to the current goal. Global replanning is used to find an alternative plan for the main teaching goal.
AHA (De Bra et al., 2000; De Bra & Ruiter, 2001; De Bra et al., 2002; De Bra et al., 2003): This is a generic system for adaptive hypermedia with the aim to bring adaptivity to all kinds of Web-based applications. The AHA supports adaptive navigation (annotation + hiding) and adaptive presentation. The general structure of AHA is similar to that of the other systems discussed above. The adaptive engine of AHA consists of three parts: a domain model, a user model, and an adaptation model. The domain model describes the teaching domain in terms of concepts, pages, and information fragments. It also contains the concepts’ relationships. The AHA uses three types of concept relationships: (a) link relationships, which represent the hypertext links among the page’s concepts; (b) generated relationships, which specify the updates to the user model related to the page’s access; and (c) requirement relationships, which define the prerequisites for the page’s concepts. The user model consists mainly of a table presenting for each page or concept an attribute value that represents how the user relates to this concept. The AHA user model differs from other systems in that the concepts’ attributes can be nonpersistent and have negative values. The adaptation model consists of a collection of rules that define the adaptive behavior of AHA. Generated rules (corresponding to the generated relationships) and requirement rules (corresponding to the requirement relationships) are part of this model.
ILESA (L pez et al., 1998a; L pez et al., 1998b): This is an intelligent learning environment for the Simplex Algorithm. It implements adaptive sequencing (lesson, problem) and provides problem-solving support. ILESA follows the traditional model of an intelligent tutoring system with six components: engine, expertise module, student diagnosis module, student interface, instructional module, and problem generator. The expertise module in ILESA is a linear programming problem solver for the Simplex algorithm. The system provides a great number of different ways to solve a problem, because this system needs to allow for the diagnosis of a student’s answers. The student diagnosis module provides a graph of learned skills. The domain is broken down into a list of skills for solving a Simplex problem, and a graph representing the relationships among skills is presented. The student model consists of an array of numbers representing the student’s score for each of the basic skills. The problem generator is used to generate an unlimited number of problems and to provide the student with the appropriate types and levels of problems. The instructional module controls the pedagogic functioning (problem posed, help offered) of the system and coordinates the actions of the expert system, the student diagnosis module, and the problem generator. The engine contains the control mechanism that guides the behavior of the system.

Client/Server Architecture for E-Learning Personalization

Most e-learning systems are implemented by using a client/server architecture. Because of the nature of the application, a client/server architecture seems to be the natural fit. In fact, an e-learning system needs to have a kind of centralized server for course management and authoring, while the clients are heavily distributed. The e-learning service or content can be distributed from a server to more clients through a network through the use of three relevant architectures:

Thin client architecture: This is a centralized managed system with applications executed on a server.
Proprietary client architecture: This is a stand-alone client application developed to support a specific service.
Internet client architecture: This is a Web-, wap-, or I-mode-browsers system.

Because of different device capabilities, different learners’ profiles and preferences and different learning strategies, an adaptation of the content and presentation are needed before they can be presented to the user. This adaptation can be done on the server (PHP, ASP, XSP), on a proxy (i.e., AvantGO), or on the client (XML/XSL or XHTML/CSS) (Butler, 2001). The main weaknesses of these solutions are as follows:

Not all browsers support content-control negotiation, so the server must make assumptions about the browser’s ability to present the content.
Heavy server-side applications may slow the server.
Often, the content is made to utilize one browser’s technological facilities application.
The lack of care given to the content/presentation layer causes problems for different-sized browsers.
The adaptation cannot be a dynamic adaptation according to the user’s profile.

< Day Day Up >