Using a Bayesian Knowledge Base for Hint Selection on Domain Specific Problems

A Bayesian Knowledge Base is a generalization of traditional Bayesian Networks where nodes or groups of nodes have independence. In this paper we describe a method of generating a Bayesian Knowledge Base from a corpus of student problem attempt data in order to automatically generate hints for new students. We further show that using problem attempt data from systems used to teach propositional logic we could successfully use the created Bayesian Knowledge Base to solve other problems. Finally, we compare this method to our previous work using Markov Decision Processes to generate hints.