Enhancing skill prediction through generalising Bayesian knowledge tracing

Learning Analytics (LA) have been widely investigated and applied to understand and optimise the learning process and environment. Among a number of LA tools, Bayesian Knowledge Tracing (BKT) was developed aiming at predicting the probability that a skill has been successfully acquired by a learner. While current development has proved BKT to be sufficiently accurate in prediction and useful, the state-of-the-art BKT methods suffer from a number of shortcomings such as the incapability to predict multiple skills learnt by a student. In this paper, we extend the ordinary BKT model to predict unlimited number of skills learned by a learner based on a non-parametric Dirichlet Process (DP). Another characteristic of our approach is that it can easily incorporate prior knowledge to our model resulting in a more accurate prediction. The extended model is more generic and able to handle border applications. We have developed two efficient approximate inference methods based on Gibbs sampling and variational methods.