Applying Learning Factors Analysis to Build Stereotypic Student Models

This paper demonstrates how stereotypic student groups can be created to enhance cognitive models in computer tutors. Computer tutors use cognitive models to track what skills students are learning and what practice attempts are most needed; often, these models contain relatively few individualized student parameters due to computational concerns. A hybrid approach emphasizing tractability and customization is used to balance the need for computable cognitive models and more flexibility to reflect learner characteristics. We use learning factors analysis to incorporate these learner characteristics and demonstrate that the resulting groups do require different cognitive models. In particular, learning rates and the actual skills that students seem to be learning based on mathematical models differ between groups.