Deep knowledge tracking and course recommendation model for online education learners

Existing course recommendation models struggle to capture the dynamic evolution of learners’ interests, while traditional knowledge tracing methods miss long-term dependencies and neglect topic features and forgetting patterns in long interaction sequences. This study improves both tasks. For course recommendation, a model integrating multi-layer attention and a gated recurrent network hierarchically fuses long- and short-term interests and models their temporal evolution. For knowledge tracing, a self-attention model with temporal enhancement is proposed, incorporating a question feature fusion module and response time intervals to simulate the forgetting curve. The recommendation model achieves HR@5, HR@10, HR@20 of 0.557, 0.684, and 0.763. The knowledge tracing model attains an AUC of 0.828 and accuracy of 0.764. In real platform deployment, the click-through rate increases to 64.8% and the course completion rate to 55.3%. The knowledge mastery prediction accuracy reaches 86.1%, with the average learner score rising to 86.5, demonstrating effective modeling for personalized online education.