A Bayesian network-based Simulation Environment for Investigating Assessment Issues in Intelligent Tutoring Systems

Abstract

Assessment is a crucial task in developing a high quality intelligent tutoring system that can provide appropriate material to learners at the right timing. In order to explore some theoretical aspects of the assessment task, we construct a simulator that creates simulated test results of students. The simulated students may make lucky guesses in responding to questions that they are not competent enough to return correct responses, but they may also make unintentional errors on questions that they should be able to handle correctly. We cope with this uncertainty problem by incorporating Bayesian networks in the simulator. This paper elaborates on details of the simulator, and overviews experimental results of some studies that we have explored with the simulator. Experience not only indicates that this simulator is instrumental for us to explore the nature of the assessment task, but also shed light on how the simulator may be improved.