Self-weighted Multi-view Fuzzy Clustering
Xiaofeng Zhu,Shichao Zhang,Yonghua Zhu,Wei Zheng,Yang Yang
DOI: https://doi.org/10.1145/3396238
IF: 4.157
2020-08-31
ACM Transactions on Knowledge Discovery from Data
Abstract:Since the data in each view may contain distinct information different from other views as well as has common information for all views in multi-view learning, many multi-view clustering methods have been designed to use these information (including the distinct information for each view and the common information for all views) to improve the clustering performance. However, previous multi-view clustering methods cannot effectively detect these information so that difficultly outputting reliable clustering models. In this article, we propose a fuzzy, sparse, and robust multi-view clustering method to consider all kinds of relations among the data (such as view importance, view stability, and view diversity), which can effectively extract both distinct information and common information as well as balance these two kinds of information. Moreover, we devise an alternating optimization algorithm to solve the resulting objective function as well as prove that our proposed algorithm achieves fast convergence. It is noteworthy that existing multi-view clustering methods only consider a part of the relations, and thus are a special case of our proposed framework. Experimental results on synthetic datasets and real datasets show that our proposed method outperforms the state-of-the-art clustering methods in terms of evaluation metrics of clustering such as clustering accuracy, normalized mutual information, purity, and adjusted rand index.
computer science, information systems, software engineering