Multi-view Clustering: A Brief Review

Abstract

With the rapid development of information technology, data exhibit multi-view characteristics, and particularly single-view data cannot comprehensively describe the information of all examples. It is very significant to sufficiently make good use of the information from different views. A good multi-view learning strategy may lead to performance improvements. Multi-view clustering is one of the important branches in multi-view learning. The key problem is how to use information effectively from multiple different views, so as to discover the underlying structure of data more accurately. In general, it uses the complementary information available to improve clustering performance in multiple views. Meanwhile, it needs to ensure that the clustering results are consistent. In this paper, we review a number of representative multi-view clustering approaches in the different fields, which can be classified into four groups: cooperative style approaches, graph-based style approaches, multiple kernel learning-based approaches, and subspace learning-based approaches. Therefore, we try to summarize and analyze the development of multi-view clustering. Finally, we point out some specific challenges that are expected to improve further research in this rapidly developing field.