Fine-Grained Question-Concept Difficulty Calibration for Knowledge Tracing

Knowledge tracing aims to trace students' changing knowledge states and predict their future answering performance by analyzing their learning trajectories. However, existing mainstream knowledge tracing models ignore the similarity and difficulty differences between knowledge concepts and questions, which makes these models unable to explore the rich information at the question level. To this end, we propose a fine-grained question-concept difficulty calibration for knowledge tracing (QCDCKT) model. By incorporating a student ability calibrating attention module with a Question-Concept matrix(Q-C matrix) and a question difficulty calibration attention module based on a LSTM model with a multi-head attention mechanism, addressing the limitations of existing methods that overlook the similarity and difficulty differences between questions and concepts. Then, we use the item response theory three-parameter(IRT3) model to predict students' responses to questions. Finally, we compare the proposed QCDCKT method with a variety of knowledge tracing models on four benchmark datasets. The experimental results demonstrate that QCDCKT outperforms the state-of-the-art methods, and the effectiveness of each module is validated.