Interaction Sequence Temporal Convolutional Based Knowledge Tracing

Knowledge tracing has become a major method for tracking changes in students' knowledge states. However, most of the existing models ignore the temporal interaction information in the learning process, resulting in insufficient interpretability. Besides, most of the models are based on RNN, which has issues with long sequence learning dependencies and gradient vanishing. In this paper, we propose an interaction sequence temporal convolutional knowledge tracing (ISTCKT) with temporal interaction model to address the above issues. We introduce three temporal factors that reflect students' learning interaction behavior and calculate the interaction temporal vector, which is concatenated with the learning interaction sequence to form the original input data. Meanwhile, we explore the deep information between exercises and knowledge concepts with graph attention networks and extract knowledge states with temporal convolutional networks, and ultimately predict the probability of students answering the next exercise correctly. We conduct extensive experiments on three public datasets to validate the effectiveness of the proposed model with better prediction performance.