Using a probabilistic student model to control problem difficulty

Bayesian networks have been used in Intelligent Tutoring Systems (ITSs) for both short-term diagnosis of students' answers and for longer-term assessment of a student's knowledge. Bayesian networks have the advantage of a firm theoretical foundation, in contrast to many existing, ad-hoc approaches, In this paper we argue that Bayesian nets can offer much more to an ITS, and we give an example of how they can be used for selecting problems. Similar approaches may be taken to automating many kinds of decision in ITSs.