Multi-View Fuzzy Clustering for Multi-Layer and Multi-Attribute Graphs

Multi-view attributed graphs (MVAGs) provide rich structural and attribute information, but existing clustering methods struggle to jointly exploit multi-view attributes and multiple graph structures. Moreover, they often focus on either visible view collaboration or hidden feature extraction, failing to capture the synergy between the two. To address these challenges, we propose ITF-MVFC (Multi-View Fuzzy Clustering with Intrinsic and Topological Features), a novel clustering model designed for MVAGs. ITF-MVFC first constructs proximity matrices from topological connections and then introduces a Double Visible-Hidden Feature Extraction (DVHFE) mechanism based on non-negative matrix factorization (NMF). This extracts both intrinsic and topological visible-hidden feature representations. To enhance sparsity and interpretability, we further employ network lasso regularization, enabling effective cooperative learning between intrinsic and topological views. Finally, a fuzzy clustering objective function is established to integrate these multi-view representations. Experiments on synthetic, real-world, and large size datasets show that ITF-MVFC consistently outperforms state-of-the-art clustering methods on both external metrics and internal validation indices across multiple datasets.