Difficulty Prediction and Individualized Matching AKT

Knowledge Tracing is a crucial task in the domain of intelligent tutoring systems, which aims to predict a student's future performance based on their past records. However, existing research fails to explicitly model the relationship between question difficulty and response prediction tasks and lack a clear modeling of individualized student features. This paper addresses these shortcomings by introducing the Difficulty Prediction and Individualized Matching AKT (DPIM-AKT) model, aimed at effectively leveraging information about question difficulty and uncovering personalized student characteristics. Specifically, we propose the Question Difficulty Prediction (DP) task to predict the difficulty of questions, enhancing the learning of question representations. We propose the Individualized Knowledge State Matching (IM) task to explicitly extract personalized features from a student's history of responses. By mining individualized patterns in a student's performance, our model achieves a more nuanced understanding of each student's learning style. Extensive experiments conducted on three real-world datasets demonstrate the superiority of DPIM-AKT over current state-of-the-art Knowledge Tracing models. Notably, DPIM-AKT also exhibits superior interpretability.