Deep multi-view clustering via structure preserving learning

doi:10.1016/j.knosys.2026.116096

Deep multi-view clustering via structure preserving learning

2026-04-30

0

PRE

AI

DOI：10.1016/j.knosys.2026.116096

Original

Original request for help

Save

Abstract

En 中文

Deep multi-view clustering has attracted considerable attention due to its ability to effectively integrate multi-view information and enhance clustering performance. First, representation learning usually relies solely on clustering pseudo-labels while overlooking the maintenance of local neighborhood relations. Second, samples from the same cluster across different views are often insufficiently aligned. Finally, during the fusion process, most approaches only adopt view-level weight allocation and fail to capture sample-level differences. To mitigate these issues, we introduce a method called deep multi-view clustering via structure preserving learning(SPLMVC). Specifically, we design an inner-view structure preserved embedding learning mechanism, which simultaneously ensures the preservation of global cluster structures and the consistency of local neighborhoods. In addition, we introduce a cross-view structure alignment learning mechanism, thereby obtaining more robust and discriminative clustering representations. Finally, we develop a sample-level adaptive weighted fusion strategy that dynamically assigns learnable weights to each sample across different views for fine-grained feature integration. Extensive experiments on multiple public datasets demonstrate the effectiveness and superiority of our method. The code of this paper is available at https://github.com/mwmytu/SPLMVC .

Keywords:

Deep multi-view clustering

structure preserving learning

representation learning

cross-view alignment

sample-level fusion

Deep multi-view clustering via structure preserving learning

Abstract

Journal

Organization