Embedded Bayesian network student models

The modeling of the student cognitive state requires to take into account uncertainty, and during the past decade the use of Bayesian networks has grown as a method for dealing with such a problem. Many different ad-hoc models have been built in user modeling as well as in student modeling, using either expert knowledge elicitation or machine learning techniques but none of these methods is perfectly adapted to the case of student modeling. Moreover, the evolution of the student cognitive state only leads to probability update in these models, whereas we think that the topology of the network should also vary in order to reflect the changes in the student knowledge structure. We propose a general framework for embedding different Bayesian network student models in an architecture that handles transitions between them and dynamic adaptation to the learner. We aim at specifying and developing an application that could provide help to build such models without having to deal with the difficulties of using belief networks.