Personalized Learning Resources Recommendation for Online Education using Hybrid Collaborative Filtering Model

The explosive growth of online education resources makes "information overload" and "learning Trek" become prominent challenges. Collaborative filtering (CF) recommendation technology is the key to meet this challenge, but when it is applied in the field of education, it faces the inherent limitations of data sparsity, cold start, interest drift and lack of domain knowledge. This study proposes a hybrid collaborative filtering model (HMD-CF) that integrates multi-source information and deep learning enhancement. The model innovatively integrates user attributes, resource content, temporal behavior and knowledge map. It uses the "double tower" architecture and attention mechanism for feature learning, and uses graph neural network (GNN) to model high-order semantic relationships. At the same time, it designs a self-adaptive cold start strategy based on meta learning. Experiments on EDX and self built K12 datasets show that HMD-CF has a high accuracy（ Precision@10 892), recall rate（ Recall@10 815) was significantly better than the baseline models such as user CF, Item CF, MF, NCF and DMF. Especially in the cold start scenario, the click through rate of new users increased by 42.7%. This system provides a personalized recommendation solution that is accurate, interpretable and suitable for complex scenes for the online education platform.