The Masked Simple Graph Contrastive Learning for Recommendation

Contrastive learning (CL) has recently sparked a productive line of research in the field of recommendation, due to its ability to extract self-supervised signals from raw data that align well with the needs of recommender systems, addressing the data sparsity issue. Graph neural networks, a deep learning algorithm for modeling graph-structured data, have proven to be efficient in recommendation systems. A typical pipeline of CLbased graph recommendation models involves first augmenting the user-item bipartite graph with structural perturbations, and then maximizing the consistency of node representations across different graph augmentations. To tackle issues of data noise and data scarcity in deep graph learning, research on graph data augmentation has intensified in recent years. However, some studies have suggested that graph augmentations were deemed necessary but played only a minor role. Drawing inspiration from these findings, this paper proposes a technique that adds uniform noise to node representations to create contrastive views, achieving strong performance on popular datasets.