Enhancing Graph-Based Recommendations via Fuzzy Clustering and Gated Multi-View Fusion

In the era of information overload, recommender systems have become indispensable for filtering and personalising content. Recent advances in Graph Convolutional Networks (GCN) have achieved state-of-the-art performance by effectively modelling high-order user-item interactions. However, data sparsity remains a critical challenge, as limited interactions can severely degrade recommendation quality. This paper proposes FuzzGCN, a novel approach that integrates fuzzy machine learning techniques into a graph-based recommendation framework. By leveraging Fuzzy C-Means (FCM) clustering to capture global item groupings and employing a gated fusion mechanism to dynamically integrate these insights with local graph propagation signals, FuzzGCN improves the robustness of item representations under sparse conditions. Moreover, we utilise contrastive learning to align the multi-view representations, further improving the model’s discriminative power. Extensive experiments on three public datasets demonstrate that FuzzGCN consistently outperforms several competitive baselines, validating its effectiveness in real-world recommendation tasks.