Multi-View Matrix Completion for Clustering with Side Information

In many clustering applications, real world data are often collected from multiple sources or with features from multiple channels. Thus, multi-view clustering has attracted much attention during the past few years. It is noteworthy that in many situations, in addition to the data samples, there is some side information describing the relation between instances, such as must-links and cannot-links. Though side information has been well exploited in single-view clustering, they have rarely been studied in multi-view scenario. Considering that matrix completion has sound theoretical properties and demonstrates excellent performance in single-view clustering, in this paper, we propose the first matrix completion based approach for multi-view clustering with side information. Instead of concatenating multiple views into a single one, we enforce the consistency of clustering results on different views as constraints for alternative optimization, and the global optimal solution is obtained since the objective function is jointly convex. The proposed Multi-View Matrix Completion (MVMC) approach exhibits impressive performance in experiments.