Dual representation learning for one-step clustering of multi-view data

Abstract

In real-world applications, multi-view data is widely available and multi-view learning is an effective method for mining multi-view data. In recent years, multi-view clustering, as an important part of multi-view learning, has been receiving more and more attention, while how to design an effective multi-view data mining method and make it more pertinent for clustering is still a challenging mission. For this purpose, a new one-step multi-view clustering method with dual representation learning is proposed in this paper. First, based on the fact that multi-view data contain both consistent knowledge between views and unique knowledge of each view, we propose a new dual representation learning method by improving the matrix factorization to explore them and to form common and specific representations. Then, we design a novel one-step multi-view clustering framework, which unifies the dual representation learning and multi-view clustering partition into one process. In this way, a mutual self-taught mechanism is developed in this framework and leads to more promising clustering performance. Finally, we also introduce the maximum entropy and orthogonal constraint to achieve optimal clustering results. Extensive experiments on seven real world multi-view datasets demonstrate the effectiveness of the proposed method.