Bayesian Cognitive State Modeling for Adaptive Serious Games

Bayesian modeling of cognitive state is one possible approach to user modeling for use with adaptivity in serious games. Adaptive educational serious games try to keep learners engaged - to keep them in the so called “Flow” channel, i.e., in the right balance between being challenged and entertained. The challenge is to intervene adaptively at the right time. The research question is when to actually adapt, and how to find quantifiable metrics for that. One way to achieve this is to model the users’ cognitive state and to adapt to high or low cognitive load, e.g., to apply dynamic difficulty adjustments. Our user modeling approach is based on Hierarchical Bayesian Models (HBM) which are suitable for drawing conclusions about the learner’s cognitive state inferred from observable variables. An important aspect is that the approach considers activity stream data such as from the Experience API (xAPI) protocol as input to achieve high interoperability and eased applicability. An evaluation with synthetic data for different user group types shows the feasibility of the approach. The model can explain differences between subjects, between subject groups and between different latent variables such as cognitive load or mental working memory capacity.