Dual consistent constraint via disentangled consistency and complementarity for multi-view clustering

Bo Li, Jing Yun

Abstract: Highlights•We propose a novel multi-view clustering framework of dual consistent constraint via disentangled consistency and complementarity for multi-view clustering (DCCMVC), which can separate the information of consistency and complementarity among multi-view data. Among them, consistency information is used for shared information consistency inference and cross-reconstruction, and complementarity information is used for within-view reconstruction.•We proposed DCCMVC implements a dual consistency constraint, which including cross-reconstruction across several viewpoints is supported by the shared information consistency inference condition, and cross-view consistency constraints supported by contrastive learning.•We validate the effectiveness of DCCMVC by conducting extensive experiments on eight datasets. The experimental results show that our method provides superior performance over several state-of-the-art multi-view clustering methods.