A Sequence Recommendation Algorithm Based on GCN Network and Self-Attention Mechanism

Traditional recommendation systems primarily rely on collaborative filtering, which struggles to effectively capture the dynamic changes in user interests. Sequential recommendation addresses this limitation by considering both user interests and social relationships to enhance recommendation outcomes. This paper proposes a sequential recommendation algorithm based on Graph Convolutional Networks (GCN) and self-attention mechanisms, termed GSRec. The algorithm utilizes GCN to extract features from both trusted and untrusted users, integrating these features with item features. Subsequently, self-attention mechanisms are employed to construct dynamic user interest features, taking into account the positive influence of trusted users and the negative influence of untrusted users. Finally, the dynamic user interest features are combined with the influence of social relationships to generate the user's future interest features.specifically, the input to the GCN includes a feature matrix and an adjacency matrix. For trusted users, the feature matrix consists of the trust levels of the top n trusted users, and the adjacency matrix represents the trust relationships among these users. For untrusted users, the feature matrix and adjacency matrix incorporate the trust and distrust levels of the top n untrusted users. Experimental results on the Epinions dataset validate the accuracy of GSRec, demonstrating its effectiveness in improving recommendation efficiency.