.png){kind=icon}
.png){kind=icon}
Return
Diffusion-driven incomplete multi-view clustering via structured regularization and semantic alignment
.png){kind=icon}
PRE
AI
Y
C
DOI：10.1007/s00530-026-02578-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Incomplete multi-view clustering is an important yet challenging task in machine learning and data mining. Due to view missingness and view heterogeneity, existing methods remain limited in missing-view recovery, cross-view semantic alignment, and clustering structure preservation, resulting in unstable clustering outcomes and constrained robustness. To address these issues, we propose a diffusion-driven incomplete multi-view clustering framework based on structured regularization and semantic alignment, termed DISCM. Specifically, DISCM first stably reconstructs missing-view features in the latent space via a diffusion-based contrastive recovery module. It then incorporates a structured regularization learning module that jointly models local neighborhood constraints and global dispersion regularization to preserve data structure and enhance inter-class separability. Next, a collaborative distribution alignment framework is introduced, which combines information synergy preservation and distribution statistical alignment to explicitly reduce semantic and distributional discrepancies between the fused representation and individual views. Finally, a clustering consistency optimization module is employed, where class-level consistency constraints and a high-confidence guidance strategy are used to jointly refine multi-view clustering predictions and progressively sharpen cluster boundaries. Extensive experiments demonstrate that DISCM significantly outperforms existing methods on multiple incomplete multi-view datasets, particularly exhibiting stronger stability and robustness under high missing-rate scenarios.
Keywords:
Incomplete multi-view clustering
Diffusion-based contrastive recovery
Structured regularization learning
Collaborative distribution alignment
Clustering consistency optimization
Journal
IF:
3.1
Papers:
2.7K
Citations:
2.7K
Organization
Cited Papers