Enhancing Recommendation Systems with Graph Contrastive Learning: A Comprehensive Study

Graph-based recommendation systems encounter multiple obstacles while offering recommendations to new users and uncommon items which gain inadequate interaction. The traditional approaches as well as deep learning models show problems when dealing with user-item sparsity which results in recommendation systems that show preference towards popular content. The research examines Graph Contrastive Learning (GCL) as a contemporary self-supervised learning approach that strengthens recommendation outcomes when dealing with difficult user and item conditions. GCL creates resilient user and item embeddings through its implementation of node feature augmentation and edge perturbation and subgraph sampling techniques while needing little labeled data. The proposed solution applies a graph contrastive loss function to accomplish three goals: it advances representation learning while removing popularity bias from recommendations and enhancing their diversity. The model receives evaluation through MovieLens 100K within a comparison framework that involves state-of-the-art GNN-based recommendation models including GCN, GAT and LightGCN. Our technique delivers substantial enhancement to recommendation precision and diversity and fairness thus proving highly beneficial for new users and unpredicted items recommendations. The research emphasizes the opportunities of contrastive learning in graph-based recommendations and it creates prospective paths toward adaptable recommendation strategies without biases.