Adaptive Bayesian Networks for Multilevel Student Modelling

In this paper we present an integrated theoretical approach for student modelling based on an Adaptive Bayesian Network. A mathematical formalization of the Adaptive Bayesian Network is provided, and new question selection criteria presented. Using this theoretical framework, a tool to assist in the diagnosis process has been implemented. This tool allows the definition of Bayesian Adaptive Tests in an easy way: the only specifications required are a curriculum-based structured domain (together with a set of weights) and a set of questions about the domain (the item pool), which will be internally converted into a Bayesian Network. In this way, we intend to make available this theoretically sound technology to educators, minimizing the knowledge engineering effort required.