Joint Robust Multi-view Spectral Clustering

Abstract

Current multi-view clustering algorithms use multistage strategies to conduct clustering, or require cluster number or similarity matrix prior, or suffer influence of irrelevant features and outliers. In this paper, we propose a Joint Robust Multi-view (JRM) spectral clustering algorithm that considers information from all views of the multi-view dataset to conduct clustering and solves the issues, such as initialization, cluster number determination, similarity measure, feature selection, and outlier reduction around clustering, in a unified way. The optimal performance could be reached when all views are considered and the separated stages are combined in a unified way. Experiments have been performed on six real-world benchmark datasets and our proposed JRM algorithm outperforms the comparison clustering algorithms in terms of two evaluation metrics for clustering algorithms including accuracy and purity.