Using Bayesian Networks to Build and Handle the Student Model in Exercise-Based Domains

Abstract

Bayesian networks are a relatively new probabilistic framework for uncertain reasoning that has been object of intense research in the last years. As a consequence, many new algorithms and techniques have been developed. In this paper we suggest how these new techniques could be used to build and handle probabilistic student models that are embedded in Intelligent Tutoring Systems for a particular class of subject domains that we have called exercise based domains. First, we make a review of the theoretical background and related work. Then, several issues related to building probabilistic student models are addressed. Specifically, how to construct the bayesian network, and how approximate and goal oriented algorithms can be used to reduce the complexity in updating the student model whenever we want to evaluate a student’s answer. Then, we discuss several applications of such student models: adaptive assessment by selecting the most informative item in terms of the sensitivity and specificity of the possible questions to ask, advancement in the curriculum, and intervention. Finally, the disadvantages of using probabilistic student models are presented together with possible ways to overcome them.