Improvement of knowledge tracing by integrating exercise features and relationships of concepts
Lige Chen, Xianzhong Liu · Journal of Physics Conference Series · 2023
Abstract Knowledge Tracing (KT) is a challenging task in personalized learning where the objective is to track the progression of students’ understanding of the concepts over time based on their learning history. Typically, there are many more exercises than knowledge points. Tracing the state of the concepts seems a reasonable approach but ignores the additional information contained in the exercise itself. Moreover, there are numerous relationships between knowledge points, and the modification of one point’s state can affect related points. In this paper, we introduce a method that splits exercises into constituent knowledge points, and incorporates the exercise features to improve the differential representation of the exercises containing the same concepts. We also describe two ways of propagating knowledge point states: one-way and two-way, which update the state of the current knowledge point and its related knowledge points. Our model is tested on two real datasets, and the experimental results show that our model outperforms the existing methods.