Student Modeling in an Intelligent Tutoring System

Abstract : This thesis explores a new approach to modeling the student in an intelligent tutoring system (ITS), by providing a student model which learns new solutions from the student. A prototype of the new approach to ITS is demonstrated in the Euclidean geometry domain. Complete C++, CLIPS, and Tcl/Tk code listings are included in the appendices for reference. Adaptable multiple software agents were targeted for implementation, based on current literature. However, the student model is found to be maintainable without multiple software agents, while still allowing for tracking several possible solution paths when monitoring student solutions. This capability contradicts previous research reported in the literature. The student model is extended by providing a learning module, which is capable of recognizing new solutions provided by the student. These new solutions may then be included in the expert knowledge base. In addition to a learning student model, other concepts from the current ITS literature are explored and implemented. Differential modeling and expectation driven analysis are analyzed, as well as the use of production rules and overlay models. Mastery levels are implemented to aid in cognitive diagnosis. Several cognitive and pedagogical concepts, such as symbolic knowledge, procedural skill, and conceptual knowledge, are explored and applied to the research. The student model prototype is both a pedagogic-content model and a subject-matter model. Additionally, a new division of labor between the student model and the instructor module in intelligent tutoring systems is described. Particularly, the student model acts strictly as a pedagogic-content model and subject-matter model, with no inferencing other than that expected of the real student. The instructor module performs all inferencing about the student's actions and knowledge.