Intelligent tutoring system based on belief networks

This paper proposes a new Intelligent Tutoring System based on Belief networks. The unique features of this system are as follows: 1. The Student model is represented by belief networks which has a Dirichret distribution as a prior, 2. The structure of the student model is constructed from the data-base by maximizing the predict distribution, and 3. Instruction strategies are described as utility functions in decision making theory. Especially, in this paper, the utility function is assumed as the expected learning effects defined by EVII (Expected Value of Instruction Information), and it is shown that it explains sufficiently teacher flexible behavior and interactive instruction process