Retrospective Evaluation of Blended User Modeling For Adaptive Educational Systems

In this paper, we are presenting a retrospective approach to evaluating user models by utilizing previously collected learning logs rather than setting up a new experiment. This approach is applied in a novel way to modeling heterogeneous types of user activity – problem solving, and browsing annotated examples. We are blending the two types of activity in the user model in an attempt to increase the accuracy of the composite model. Obtained results suggest that such blending, in fact, does make a difference both for users individually and on a global scale.