Student Model

 < Day Day Up > 



With the advance in artificial intelligence techniques and the power of computers, modeling student behaviors has become possible for inclusion into intelligent tutoring systems. The modeling process allows the system to predict the learning behavior of individual students and diagnose the causes of errors (Dede, 1986). However, to do this, one needs a model of the learner, which represents cognitive processes (e.g., information processing, analysis, synthesis, information retrieval, calculation, and problem solving); metacognitive strategies (e.g., learning from errors, when to access more information, when learning outcome has been reached); and psychological attributes (e.g., developmental level, learning style, motivational level, and interests).

The student model develops hypotheses about the student’s misconceptions and inferior performance strategies so that the pedagogical module can point them out, indicate why they are wrong, and suggest corrections (Tennyson & Park, 1987). For example, for problems that require procedural skills, mistakes are usually due to the correct execution of an incorrect procedure. The pedagogical module must identify this inappropriate strategy and take corrective action. Major information sources for maintaining the student model are (a) direct questions asked of the student, (b) student problem-solving behavior observed by the system, (c) assumptions based on the student’s learning experience, and (d) assumptions based on some measure of difficulty of the subject materials (Clancey, Barnett, & Cohen, 1982). The method for forming a student model should be based on the methods an effective human tutor uses in a one-to-one environment. Ally (2000) identified the tasks that experienced tutors use in a one-to-one tutoring situation. These tasks could be used to help develop an effective student model and pedagogical module for intelligent tutoring systems for distributed learning.

The student model should identify the student’s learning style and preference based on interaction with the student and should then send the information to the pedagogical module, which then adapts and prescribes activities to meet the needs of the student. Previous intelligent tutoring systems ignored many potentially important student variables in the diagnostic and prescriptive process by relying solely on the student’s response pattern. In most learning situations, a reliable pattern of the student’s response is not developed until the student makes significant progress on the given task. Frequently, the interactions between the system and the student required to learn a given task are not long enough to observe the demonstration of the student response pattern. A student model should provide an explicit representation of the student’s incorrect versions of the reasoning process so that remedial actions can be taken (Wenger, 1987). According to Wenger (1987), a complete intelligent tutoring system should include important learner variables at the following levels in the student modeling process:

  1. The behavioral level: This level deals with behavior and the product of behavior, without trying to perceive the knowledge state involved in its generation (Wenger, 1987).

  2. The epistemic level: This deals with the student’s knowledge state, including aspects of both the domain (general model) and strategic knowledge (inference procedure) (Wenger, 1987).

  3. The individual level (Wenger, 1987):

    (a) Architectural: A knowledge state must be represented within a cognitive architecture.

    (b) Learning: A learning model could be included in the student model to follow or anticipate the student’s acquisition of knowledge.

    (c) Stereotypical: Personality traits, learning style, and individual preferences can be important in selecting topics and presentation styles (e.g., preference for a visual or textual display, use of organizational devices, etc.).

    (d) Motivational: Levels of interest, overload, or fatigue may require temporary changes in instructional style and pace.

    (e) Circumstantial: It may be useful to understand the influence of the environment on the signals the system receives from the student. This includes some special events that distract the student or otherwise modify his or her performance, such as intervention from a teacher or the help of a friend.

    (f) Intentional: This includes the viewpoint of the student, not with respect to the subject matter, but with respect to the meaning of the tutorial interaction.

    (g) Reflexive: The model the student has in the context of a domain and of an instructional interaction.

    (h) Reciprocal: The student’s reaction to the system.

Intelligent tutoring system research has been concerned with the model that the system constructs of the student, but hardly with the model the student forms of the system. Aspy and Roebuck (1977) mentioned that “Kids don’t learn from people they don’t like.” The same may apply for intelligent tutoring systems. “People don’t learn from intelligent tutoring systems they don’t like.” As the relationship between a student and a tutoring system begins to be viewed as one of communication, the significance of reciprocal models will need to be identified, and the student’s input will need to be interpreted in light of this type of information. Hence, the system must be dynamically adapted to meet the needs and styles of learners to engage learners in the learning process.

How Student Knowledge is Represented in the Student Model

The student model is used to assess a student’s knowledge state and to make hypotheses about the student conceptions and reasoning strategies. Modeling student’s knowledge and learning behavior uses two methods. (a) Charting within the knowledge structure network those areas, which the student has mastered or has attempted to learn. (b) Applying pattern recognitions to the student’s response history for making inferences about his or her understanding of the skill and the reasoning process used to derive the response. Most intelligent tutoring systems represent the student’s knowledge state as a subset of an expert’s knowledge base. The student model is constructed by comparing the student’s performance to the computer-based expert’s behavior on the same task. Goldstein (1982) referred to this technique as the “overlay model.” In the overlay model, implicit evidence is derived from a comparison of the student’s behavior with the expert’s decision. It requires the ability to relate a given behavior to a specific set of skills (Wenger, 1987).

However, there are two limitations of the overlay model. First, the overlay model will fail if the domain allows for multiple problem-solving paradigms and if the student follows one that is not in the expertise module, possibly making good moves that the expert system would consider to be nonoptimal (Wenger, 1987). For example, a student could be using a personal metacognitive strategy that is effective for the student but is not in the intelligent tutoring system. This implies that expertise other than that of an expert for solving a problem should also be included in an intelligent tutoring system. Correct procedures used by novices and semi-experts should be included to teach and tutor learners who tend to be novice in the field. The second limitation is that the overlay modeling paradigm is usually used to evaluate the acquisition of factual information. However, errors in procedural knowledge are more complex than gaps in factual, or declarative, knowledge. The overlay model assumes that low-level behavior is caused only by the insufficient mastery of individual skills, and that distortions in the correct skills do not occur (Wenger, 1987).

To obtain an accurate student model, all of the expert’s knowledge, including first principles knowledge that the expert uses to solve problems, should be included in intelligent tutoring systems. However, experts tend not to use first principles knowledge to solve problems (Wenger, 1987). With experience, the experts’ first principles knowledge is compiled into large chunks and, hence, has lost its original form. Novices and semi-experts are more likely to use first principles knowledge. As a result, novices as well as semi-experts should be included in the knowledge elicitation process to build expert systems.

Another technique for forming a student model is to represent the student’s mislearned knowledge as variants of the expert’s knowledge. This technique is referred to as the “buggy model” (Brown & Burton, 1978). The “buggy model” may represent domain knowledge as rules and potential misconceptions as variants of the rules. In the buggy model, a diagnostic modeling scheme based on a procedural network contains all the necessary subskills for the global skill, as well as the possible buggy variants (or bugs) of each subskill. It can replace an individual subskill in the procedural network by one of its bugs, and thus attempt to reproduce a student’s incorrect behavior (Wenger, 1987). Bugs originate mostly in mislearning and forgetting which is caused by the student’s failure to follow an instructional sequence and to organize and retain information internally, or in the teacher’s failure to present sufficiently complete and unambiguous information.

There are two limitations of the buggy model. First, the modeling capability requires a proper decomposition of the skill to a minute level at which each single bug can be isolated as a variant of a separate procedure. Second, the buggy procedure network is able to identify bugs in the student’s procedure, but it does not attempt to explain them (Wenger, 1987). A student model should give an explanation about the line of reasoning when an error has been committed. This would allow the system to trace back to the original learning process or to deeper misconceptions either within the domain or in a related area. The line of reasoning of the learner would allow the system to suggest specific remediation, which is essential for effective tutoring (Wenger, 1987).

If an intelligent tutoring system can compare the student’s understanding of the subject to its own expert representation, then instruction can be organized on transferring the missing or distorted portions of the know1edge base to the student. However, this may not be realistic, because the expert’s line of reasoning is different from that of a novice. What is needed is the line of reasoning of the successful novice and semi-expert who has the same background of the learner with the problem. As the learner responds to questions and situations, conclusions about the learner’s understanding can be made. The accuracy of these conclusions can be tested by comparing them to those that the domain module would generate given only the knowledge in the student subset (Dede, 1987).

In the differential model, the intelligent tutoring system abstracts and summarizes the student’s behavior in instructional learning situations. It compares the skills demonstrated to the domain module’s responses under identical situations. The parts of the domain that the student does not understand can be determined through inference from differential weaknesses between student and expert.

After the system has developed a model of the learner, the pedagogical module then has to decide the next step. The pedagogical module is examined next.



 < Day Day Up > 



Designing Distributed Environments with Intelligent Software Agents
Designing Distributed Learning Environments with Intelligent Software Agents
ISBN: 1591405009
EAN: 2147483647
Year: 2003
Pages: 121

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net