Deep Knowledge Tracking Model Integrating Multiple Feature Personalization Factors
Yukun Wang, Xu Xingjian, MENG Fanjun · 2024
Knowledge tracing is crucial for intelligent educational systems, as it captures students' mastery of knowledge points and predicts future performance. Current methods often overlook forgetting rates and fail to integrate personalized factors from both learners and problem contexts. To address these limitations, we propose a deep knowledge tracing model that incorporates multiple personalized features. We use the Rasch model for regularization of knowledge point and problem embeddings, capturing individual differences within the same domain. A hierarchical convolutional network extracts learners' personalized learning rates and prior knowledge levels while computing forgetting rates. We also simulate interactions between learners' knowledge states and problem contexts using vector dot products, adjusting weights through a personalized knowledge-aware attention mechanism. Experiments on three real-world online education datasets show that our method outperforms existing knowledge tracing models in prediction accuracy and enhances performance through personalized design. Visual analyses of specific cases further confirm the effectiveness of our approach.