Self-Supervised Graph Attention Collaborative Filtering for Recommendation

Due to the complementary nature of graph neural networks and structured data in recommendations, recommendation systems using network techniques have become mainstream. However, there are still problems, such as sparse supervised signals interaction noise, task. Therefore, this paper proposes a self-supervised attention collaborative filtering for (SGACF). The correlation between adjacent nodes is deeply mined multi-head obtain accurate node representations. It worth noting that learning brought an auxiliary task recommendation, where supervision main assists model training tasks. A multi-view generated by data-augmentation method. We maximize consistency its different views compared same minimize other nodes. In paper, effectiveness method illustrated abundant experiments on three public datasets. results show significant improvement accuracy long-tail item robustness model.