Graph Structure Awareness and Edge Attention Collaborative Filtering

Neural networks on graphs are effective algorithms for solving the graph recommendation problem. Graph contrastive learning algorithms are effective in addressing the problem of supervised label scarcity by enhancing user and item representations. Although there are many methods to prove their effects, there are still some key problems to be solved. First, most contrastive learning methods are based on user clustering or node perturbation. The contrastive view generated by this random augmentation and heuristic method is apt to lose its semantic structure. These methods are easily influenced by noisy information when adopting the above data augmentation. Based on the problems mentioned above, we analyze them and propose a collaborative filtering framework based on graph structure awareness and the edge attention mechanism.