TSCD: Temporal Similarity-aware Cognitive Diagnosis Model

Along with the evolution of intelligent education systems, cognitive diagnosis has become an essential tool for understanding and predicting student learning behaviors. This study introduces TSCD (Temporal Similarity-aware Cognitive Diagnosis), an innovative framework that advances cognitive diagnosis through synergistic integration of temporal dynamics modeling and similarity-aware feature engineering. TSCD constructs a comprehensive feature vector that combines student embeddings, exercise difficulty, discrimination, and a newly introduced student-knowledge similarity feature. By utilizing Long Short-Term Memory (LSTM) networks, which enables the model to capture the temporal dependencies within students’ learning sequences and dynamically models their evolving knowledge states. Including the similarity feature further strengthens the model’s ability to capture the relationships between students and knowledge points, offering a richer representation of learning interactions. Experimental results on benchmark datasets show that TSCD outperforms traditional cognitive diagnosis models in predicting student performance, providing valuable insights for personalized learning and educational interventions.