.png){kind=icon}
.png){kind=icon}
Return
Projection-based incomplete multi-view consensus bipartite graph representation learning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131766.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Incomplete Multi-View Clustering (IMVC) has garnered significant attention in recent years due to its ability to effectively handle incomplete multi-view data. Although graph-based IMVC methods are widely adopted, the reliability of the obtained affinity graphs is often compromised by the presence of missing instances. Moreover, imputation-based methods are frequently adversely affected by redundant features or noise, which can distort data relationships and structures, ultimately leading to inaccurate clustering results and limiting their practical applicability. To address these challenges, we propose a novel method termed Projection-based Incomplete Multi-view Consensus Bipartite Graph Representation Learning (PIMV_CBG). Specifically, we integrate projection learning with consensus bipartite graph construction into a unified framework, where they mutually enhance each other to reduce the impact of noise and redundancy while strengthening cross-view interactions, leading to high-quality bipartite graph representations. Furthermore, imputation is performed in the clean subspace, regularized by graph constraints derived from the consensus structure, which improves imputation accuracy and fully exploits latent information. Extensive experiments on benchmark datasets have demonstrated the effectiveness of PIMV_CBG in addressing IMVC tasks, achieving state-of-the-art clustering performance in most cases. The source code is publicly available at https://github.com/superkeranbing/PIMV_CBG/tree/main .
Keywords:
Multiview learning
Clustering
Matrix factorization
Bipartite graph