Semantic View Evolution for Incomplete Multi-View Clustering

Incomplete multi-view clustering faces challenges from missing views and cross-view semantic redundancy. Existing methods typically perform information fusion within a fixed view space, making it difficult to continuously explore complementary semantics. To address this issue, we propose a Semantic View Evolution framework for Incomplete Multi-View Clustering (SVE_IMVC). SVE_IMVC first learns initial semantic representations through cross-view mutual information learning, and then introduces View Expansion and View Contraction to explore complementary semantics and suppress redundant representations, respectively, enabling the semantic view space to evolve dynamically. Based on the evolved global semantic representation, missing views are further recovered through semantic-guided completion, and the completed representations are used for final clustering. Experimental results demonstrate that SVE_IMVC effectively improves clustering performance on incomplete multi-view data.