Student Models for Prior Knowledge Estimation.

Intelligent behavior of adaptive educational systems is basedon student models. Most research in student modeling focuses onstudent learning (acquisition of skills). We ocus on priorknowledge, which gets much less attention in modeling and yetcan be highly varied and have important consequences for theuse of educational systems. We describe several models forprior knowledge estimation – the Elo rating system, itsBayesian extension, a hierarchical model, and a networked model(multivariate Elo). We evaluate their performance on data fromapplication for learning geography, which is a typical casewith highly varied prior knowledge. The result show that thebasic Elo rating system provides good prediction accuracy. Morecomplex models do improve predictions, but only slightly andtheir main purpose is in additional information about studentsand a domain.