sequential recommendation

user item

collaborative filtering

Collaborative Filtering (CF) has emerged as fundamental paradigms for\nparameterizing users and items into latent representation space, with their\ncorrelative patterns from interaction data. Among various CF techniques, the\ndevelopment of GNN-based recommender systems, e.g., PinSage and LightGCN, has\noffered the state-of-the-art performance. However, two key challenges have not\nbeen well explored in existing solutions: i) The over-smoothing effect with\ndeeper graph-based CF architecture, may cause the indistinguishable user\nrepresentations and degradation of recommendation results. ii) The supervision\nsignals (i.e., user-item interactions) are usually scarce and skewed\ndistributed in reality, which limits the representation power of CF paradigms.\nTo tackle these challenges, we propose a new self-supervised recommendation\nframework Hypergraph Contrastive Collaborative Filtering (HCCF) to jointly\ncapture local and global collaborative relations with a hypergraph-enhanced\ncross-view contrastive learning architecture. In particular, the designed\nhypergraph structure learning enhances the discrimination ability of GNN-based\nCF paradigm, so as to comprehensively capture the complex high-order\ndependencies among users. Additionally, our HCCF model effectively integrates\nthe hypergraph structure encoding with self-supervised learning to reinforce\nthe representation quality of recommender systems, based on the\nhypergraph-enhanced self-discrimination. Extensive experiments on three\nbenchmark datasets demonstrate the superiority of our model over various\nstate-of-the-art recommendation methods, and the robustness against sparse user\ninteraction data. Our model implementation codes are available at\nhttps://github.com/akaxlh/HCCF.\n