Using POMDP in Building an Adaptive Intelligent Tutoring System

An intelligent tutoring system (ITS) can teach students in a one-to-one, interactive way. It may help students achieve their learning goals better than classroom lecturing. An ITS should be able to teach adaptively based on knowledge states of students. Uncertainty is a challenge in developing an ITS. In practical tutoring, student information available to a teacher may be incomplete and uncertain. The partially observable Markov decision process (POMDP) model provides useful tools for handling uncertainty. It enables an ITS to take optimal teaching actions even when uncertainty exists in tutoring processes. In this paper, we reported an experimental ITS developed on the POMDP model. We describe the definitions of states, actions, observations in the POMDP framework, and the techniques for dealing with exponential state space and POMDP solving, which are major barriers in building POMDP based ITSs for practical applications. keywords: Intelligent system, computer supported education, partially observable Markov decision process.