One-step multi-view clustering guided by weakened view-specific distribution

In recent years, multi-view clustering has demonstrated superior performance by capturing complementary information from diverse views, garnering considerable attention. Although existing methods have shown promising results, they encounter some challenges. Many existing methods are based on view-level fusion, which disregards the sample-level global structural relationships and leads to potential misclustering of some samples. On the other hand, some one-step methods use view-specific information that contains misleading content as guidance, which leads to potential clustering bias. To address these challenges, we proposed one-step multi-view clustering guided by weakened view-specific distribution (OSMVC). First, the global structural relationships fusion clustering (GSRFC) module fuses multi-view information by capturing the sample-level global structural relationships while achieving steady cluster assignments. Second, the weakened view-specific distribution is employed as a soft distribution to guide the feature fusion and clustering process. Extensive experiments validate the effectiveness of our proposed module and method. Compared with eight state-of-the-art methods on nine datasets with different scales, OSMVC gains superior clustering performance. Our code is released on https://github.com/ykxhs/OSMVC.