Publications

Abstract : An Intelligent Tutoring System (ITS) supplements traditional learning methods and is used for personalized learning purposes that range from exploring simple examples to understanding intricate problems. The Bayesian Knowledge Tracing (BKT) model is an established method for student modeling. A recent enhancement to the BKT model is the BKT-PPS (Prior Per Student) which introduces a prior learnt for each student. Although this method demonstrates improved prediction results compared to the others, there are several aspects that limit its usefulness; (a) for a student, the prior learning is common for all skills, however in reality, it varies for each skill (b) Different students have varying learning capabilities; therefore these students cannot be considered as a homogenous group. In this paper, we aim to improve the prediction of student performance using an enhanced BKT model called the PC-BKT (Personalized & Clustered) with individual priors for each student and skill, and dynamic clustering of students based on changing learning ability. We evaluate the predictions in terms of future performance within ASSISTments intelligent tutoring dataset using over 240,000 log data and show that our models increase the accuracy of student prediction in both the general and the cold start problem.