Data Augmentation and Feature Fusion to Improve Graph Contrastive Learning for Recommender Systems

Graph Contrastive Learning (GCL) is increasingly recognized for its application in recommendation systems, primarily for obtaining higher-quality node embeddings of users and items. However, existing GCL methods mostly use random heuristic data augmentation, which may not ensure data diversity and generalization capability. To address this issue, we propose a new method named DFGCL (Data Augmentation and Feature Fusion to Improve Graph Contrastive Learning for Recommender Systems). DFGCL employs community edge reduction technology to remove non-essential edge information, thus obtaining better augmented samples, and uses approximate singular value decomposition technology to capture global structural information. Additionally, the feature fusion strategy integrates feature information from different graph-augmented samples, balancing user-item feature representations across different augmentation views, thereby enhancing the model's understanding of complex user behavior patterns. Extensive experiments on three public datasets demonstrate that our model outperforms other baseline models.