GACS: Graph Contrastive Learning with Augmented Collaborative Signals for Recommendation

Graph Contrastive Learning (GCL) is a self-supervised learning paradigm that extracts features from unlabeled data by comparing similarities and differences between views and has shown great potential in graph-based recommender systems. Most GCL-based recommender models construct distinguishable contrast views through graph perturbation that makes embeddings uniformly distributed, which alleviates the over-smoothing problem. However, the perturbation obscures the collaborative signals between entities, which reduces the representation ability of the model. Additionally, current GCL-based recommender models utilize invariant weights to optimize contrastive learning, recommendation, and other tasks, which cannot adapt to the dynamic changes of tasks during the training phase. To address these issues, we propose a Graph contrastive learning model with Augmented Collaborative Signals for recommendation (GACS). GACS constructs the differentiation enhancement view by adding noise to the embeddings and the collaborative signals augmentation view by leveraging implicit relationships between entities. Both views are used for contrastive learning to mitigate the over-smoothing problem and enhance the representation ability. Furthermore, GACS employs learnable weights to optimize tasks dynamically and improves the robustness through parameter limitations. Extensive experiments demonstrate the effectiveness and rationality of our model. The source code of our model is available at https://github.com/CoderYu233/GACS .