Incomplete multi-view clustering based on information fusion with self-supervised learning

Clustering algorithms aim to analyze data structures and properties, grouping the data based on their underlying structural characteristics. Traditional multi-view clustering algorithms focus on combining data shared by multiple views to perform cluster analysis, and such algorithms typically limit the completeness of the data in each view. In real-world applications, it is common that the samples within each view are missing, which reduces clustering performance. In this paper, we propose a novel incomplete multi-view clustering algorithm that addresses the sample-missing problem by leveraging self-supervised information fusion to integrate both global and local information. Data pair construction, global information extraction, and missing information completion are three core modules for the proposed algorithm: (1) constructing multi- view positive and negative pairs through contrastive learning to identify the differences between data samples in a self-supervised manner; (2) forming adjacency matrices to capture the manifold structure of the entire sample data and introducing spectral loss to tighten the representation of similar clusters in the feature layer; and (3) employing the kernel regression method to estimate the missing information in a nonlinear manner by calculating the cross-view metric relationship between the existing data and the missing data, thereby improving data integrity. The proposed network framework combines local information complementation and global information extraction. The experimental results show that our proposed method outperforms existing state-of-the-art methods, achieving an average performance gain of 4.75% accuracy across multiple datasets.