A Time-Sensitive Knowledge Tracing Method for Educational Data Analysis

Knowledge tracing aims to dynamically model students’ knowledge states by analyzing their learning interaction behaviors and represents an important problem in educational data analysis and learning analytics. With the rapid growth of educational data, learning behaviors exhibit pronounced temporal characteristics and irregular time intervals. Effectively modeling the influence of learning order and temporal factors on knowledge retention remains a key challenge in knowledge tracing research. To address this challenge, this paper proposes a time-sensitive knowledge tracing method for educational data analysis, termed Ts-DKT. The proposed method is built upon the Transformer architecture and incorporates a hybrid positional encoding mechanism to enhance sensitivity to learning sequence order. In addition, a time-decay mechanism is introduced to model the forgetting process of knowledge over time, enabling a more accurate representation of the dynamic evolution of students’ knowledge states. Furthermore, a self-attention mechanism is employed to capture long-range dependencies within learning interaction sequences. Experimental results on multiple real-world educational datasets demonstrate that Ts-DKT consistently outperforms representative knowledge tracing models, including DKT, BKT, and AKT, across evaluation metrics such as AUC, accuracy, precision, and recall. These findings indicate that the proposed method achieves improved predictive performance and robustness in educational knowledge tracing tasks, providing valuable algorithmic insights for educational data analysis and learning analytics.