In business analysis, providing effective recommendations is crucial for boosting company profits. Graph structures, especially bipartite graphs, are favored for analyzing complex data relationships. Link prediction is crucial for recommending specific items to users. Traditional methods have primarily focused on binary classification tasks. These methods, which identify patterns in graph structures or use representation techniques like graph neural networks (GNNs), face challenges with increasing data volume and label count. Data growth strains system performance and efficiency. More labels intensify data sparsity, as users and items focus on only a few labels, leading to sparse matrices that hamper recommendation algorithms. To tackle these issues, we introduce the Graph Contrastive Learning for Multi-label Classification (MCGCL) model. It uses contrastive learning to improve recommendations and has two training phases: a main task of holistic user–item graph learning to grasp user–item relationships, and a subtask of constructing homogeneous user–user (item–item) subgraphs to capture user–user and item–item relationships. Comparative experiments with state-of-the-art methods confirm the effectiveness of MCGCL, highlighting its potential for improving recommendation systems.