Attention-Aware Graph Contrastive Learning for Movie Recommendation Services

Graph-based recommendation systems (RSs) have received significant attention because of their ability to capture user-item interactions more effectively. However, existing Graph Neural Network (GNN)-based recommenders often struggle with data sparsity and lack effective mechanisms for learning robust representations. To address these limitations, we propose Attention-Aware Graph Contrastive Learning (AGCL), a novel self-supervised framework that integrates attention mechanisms within contrastive learning to enhance feature representations. AGCL employs adaptive graph augmentations, including edge perturbation, feature masking, and subgraph sampling, to generate diverse contrastive views of the user-item interaction graph. The attention mechanism ensures that informative interactions are prioritized, improving the model’s ability to distinguish relevant patterns. The experiments on benchmark movie recommendation datasets, MovieLens and the Netflix Prize, have demonstrated that the proposed AGCL outperforms state-of-the-art GNN-based and contrastive learning-based models in both standard and cold-start scenarios. Our ablation experiments further highlight the effectiveness of attention-aware contrastive learning in mitigating data sparsity and enhancing the robustness of recommendations.