A multi-view ensemble clustering approach using joint entropy

Abstract

Multi-view clustering can capture complementary and consistent information from different views, which is a core research topic in the fields of machine learning and pattern recognition. However, existing multi-view clustering methods ignore the quality of each base clustering cluster, which may affect the clustering performance. In this study, a multi-view ensemble clustering approach is proposed using joint entropy to evaluate the base clustering clusters. Firstly, an uncertainty index of base clustering clusters is defined using joint entropy which characterizes the importance and quality of each cluster. Secondly, a weighted co-association matrix is constructed using the uncertainty index, and useless entries are removed from the matrix, making the co-association matrix more reasonable. Thirdly, a candidate clusters selection strategy based on the stability index and a Multi-view Ensemble Clustering (MvEC-DoS) algorithm is proposed. In the end, experiments on five benchmark datasets validate the efficacy of our approach compared to other state-of-the-art methods.