A Graph Contrastive Learning with Feature Perturbation for Recommender Systems

This paper focuses on the efficient mining of user-item relationships and the aggregation of neighborhood information in bipartite graph-based recommender systems and proposes a Graph Contrastive Learning recommendation model with Feature Perturbation (GCL-FP), utilizing a graph convolutional encoder with a residual structure to alleviate the over-smoothing problem of node representation.