Enhancing Social Recommendation with Global Dependency Modeling base on Self-Attention Graph Neural Network

Social recommendation aims to provide personalized content recommendations by leveraging user relationships and interactions within a social network. It commonly employs techniques like collaborative filtering and graph neural networks. However, traditional graph neural networks often encounter limitations when dealing with social network data. They primarily rely on local information propagation and consider only immediate neighboring nodes, making it challenging to capture global dependencies comprehensively within the social network. In our research, we present a novel social recommendation model that extends the traditional graph neural network by incorporating self-attention mechanisms. This addition allows the model to better capture global dependencies between nodes. By considering relationships among all nodes in the graph, this module can identify long-range dependencies between nodes, providing more comprehensive information. This innovative model effectively enhances the performance of social recommendation, resulting in more personalized and relevant content recommendations for users.