Educational resources recommendation algorithm based on GMF-MLP-NeuMF prediction model

This study proposes a hybrid recommendation model (GMF-MLP-NeuMF) enhanced with knowledge graphs and temporal attention mechanisms to improve educational resource recommendation systems. Traditional collaborative filtering methods often struggle with precision in modeling user-resource interactions, particularly in dynamic educational environments. To solve this problem, this paper combines generalized matrix decomposition (GMF) and multi-layer perceptron (MLP), and introduces knowledge graph and temporal attention module to construct a comprehensive recommendation system. The system can capture the dynamic interest change of users and effectively deal with the cold start problem of educational resources. Through the experimental results, the proposed model has achieved significant improvement in the traditional recommended performance indicators (AUC, F1 score, etc.). Specifically, the AUC value of KG-TNeuMF model was 0.812, F1 score was 0.643, and NDCG@10 was 0.478, which significantly improved the recommendation accuracy compared with the traditional NeuMF model (AUC: 0.782, F1: 0.614, NDCG@10: 0.432). In addition, knowledge knowledge coverage (KCR) and learning path fit (LPC) of KG-TNeuMF model were 0.76 and 0.83 respectively, which were significantly higher than other models. These results show that the proposed model has great potential for application in practical education platforms. These advancements offering insights for optimizing IT-enabled educational resource allocation.