A graph neural network recommendation algorithm based on multi-scale attention and contrastive learning

When facing sparse user-item interaction data, recommendation systems often struggle to learn high-quality representations, which in turn affects the recommendation performance. To address this issue, this paper proposes a graph neural network-based recommendation algorithm with multi-scale attention and contrastive learning (GR-MC). First, a dedicated graph structure augmentation strategy based on user-focused edge dropout is designed to intentionally reduce the dominance of high-degree user nodes in neighbor aggregation, effectively alleviating degree bias and improving the model's generalization ability. Second, a multi-scale attention embedding propagation mechanism is proposed to enhance the modeling of higher-order neighbor relationships. Finally, by treating node self-discrimination as a self-supervised task, contrastive learning is introduced to provide auxiliary signals for representation learning, thereby improving the discriminative ability of embeddings and the robustness of the model. Experimental results show that GR-MC outperforms existing methods on multiple public datasets, especially on the highly sparse Amazon-book dataset, where Recall@20 improves by 24.69%, fully demonstrating its effectiveness and robustness in sparse environments.