Bipartite graph-based multi-view clustering conducts clustering of samples in accordance with the relationships between samples and anchors, and has demonstrated significant advancements in recent years. Predefined bipartite graphs with fixed anchors may not reflect the underlying clustering structure accurately, leading to the degradation of clustering performance.
To solve the problems, a research team led by Yuxin HUO published their new research on 15 September 2026 in Frontiers of Computer Science.
The team proposed a Structure Sparsity-Induced Bipartite Graph (SSBG) learning method to dynamically construct view-specific bipartite graphs with automatically learned anchors. On one hand, representative anchors that capture the underlying data structure are identified via a structural sparsity-induced selection matrix. On the other hand, both view-specific characteristics and cross-view consistency are effectively preserved through a self-representation-based reconstruction mechanism. Compared with the existing research results, the proposed method has advantages in clustering performance.
In the research, key samples are selected by multiplying the feature matrix of each view with a common selective matrix with structure sparsity induced by the -norm. Meanwhile, view-specific bipartite graphs are learned jointly by minimizing the reconstruction error within the self-representation framework. Benefiting from the superior performance of the self-representation-based subspace clustering model, the learned bipartite graphs also possess excellent capabilities in characterizing clustering structures. Finally, these view-specific bipartite graphs are fused to learn a consistent one, which contains the common and complementary information from multiple views.
The clustering partition can be directly obtained from the connected components of the consistent bipartite graph by employing the Laplacian rank constraint. Experimental results verify the advantages of the proposed method over state-of-the-art methods.
Future work can focus on constructing view-specific anchors and exploring high-order correlations among views in the fusion process.
DOI:10.1007/s11704-025-50224-3