Incorporating forgetting in the Personalized, Clustered, Bayesian Knowledge Tracing (PC-BKT) model

Personalization and adaptation are at the core of Intelligent Tutoring Systems. The Bayesian Knowledge Tracing (BKT) Student Model is a time-tested method that maintains information about students´ knowledge levels for the different skills in the topic domain. In our previous work, we had proposed the Personalized, Clustered, Bayesian Knowledge Tracing (PC-BKT) model that individualizes the learning of skills for each student and additionally improves the prediction for the cold start problem. A clustering of both students and skills based on a student and skill capability matrix was used to learn the prior skills to deal with the cold start problem, which is the prediction for either new skills or new students. Both the BKT and the PC-BKT models assume that a skill once learnt is never forgotten. But forgetting is pervasive. If a previously learnt skill is not used for a while, there is a higher chance of forgetting it. One of the factors that influence the forgetting is the time duration before the current attempt at using a skill and the previous attempt. We incorporate forgetting as a time decay function in the BKT and PC-BKT models and show significant increase in the accuracy of the student prediction.