A Temporal-Enhanced Model for Knowledge Tracing

Knowledge Tracing (KT) aims to predict students' future practice performance through their historical interaction with Intelligent Tutoring Systems (ITS). This method plays an important role in computer-assisted education and adaptive learning research. In the learner's learning process, as the learning time increases, the time distance between the learner's historical records continues to increase, resulting in a long-term dependency problem when capturing the correlation of knowledge concepts in exercises. In addition, the learner's learning status is affected by time. How to accurately judge the learner's knowledge status at different time levels is also a challenge. To tackle the above problems, we propose A Temporal-Enhanced Model for Knowledge Tracing (TEKT). On the one hand, the problem of long-term dependence is solved by using the Probabilistic Sparse Attention mechanism; On the other hand, the Fine-grained Temporal Features are embedded to capture learner's knowledge status at different time granularities. The method proposed in this paper has been fully experimented and verified on three datasets. The experimental results show that the proposed method demonstrates an improvement in Accuracy (ACC) and Area Under The Receiver Operating Characteristic Curve (AUC) evaluation metrics, which compared to the existing KT methods. Thus, it proves the effectiveness of the proposed method.