In the field of movie recommendation, collaborative filtering algorithms and content filtering techniques have long been traditional methods adopted. However, these methods often face challenges such as data sparsity, difficulty in cold start, and difficulty in fully capturing the complex interaction relationships between users and items. In response to the above issues, this study proposes a movie recommendation system based on Heterogeneous Graph Neural Network (HGNN). The system first constructs an interaction diagram that integrates five entities: users, movies, directors, actors, and genres. Heterogeneous graph structures help to integrate multiple associations more comprehensively. Through the neighborhood aggregation mechanism of graph neural networks, the system can deeply explore the deep connections between users, movies, and their attributes, thereby effectively achieving collaborative modeling of multi entity relationships. The experiment was validated using the MovieLens-25M dataset. The results show that compared with traditional collaborative filtering methods (such as UserCF), matrix factorization techniques (such as NeuMF), and isomorphic GNN models (such as GraphSAGE), the system proposed in this paper has significant improvements in multiple evaluation metrics: Recall@20 An increase of 18.7%, NDCG@20 An increase of 15.2% and a coverage rate increase of 22.3%. Especially in cold start scenarios, the improvement in recommendation accuracy is more prominent, reaching 31.5%.
Bao et al. (Thu,) studied this question.