Research on Product Recommendation Algorithm Based on Neural Network for User Interests

In view of the shortcomings of traditional product recommendation algorithms in mining user interests, we proposes a new recommendation algorithm based on heterogeneous graph neural networks. First, a user-item-tag heterogeneous graph structure is constructed, fully considering the diverse relationships between users, items, and tags. Secondly, the HetGNN heterogeneous graph neural network model is introduced to effectively integrate information between different types of nodes and capture the user's interest characteristics more comprehensively. Furthermore, by learning the user's representation on the heterogeneous graph, personalized product recommendation is achieved. Experimental results show that compared with traditional methods, this algorithm has significantly improved recommendation accuracy and degree of personalization, proving the effectiveness and superiority of heterogeneous graph neural networks in user interest mining. Future research can further optimize the algorithm, consider more data sources and features, and improve the performance and user satisfaction of the recommendation system.