Multiview clustering via consistent and specific nonnegative matrix factorization with graph regularization

Multiview clustering is a hot research topic in machine learning and computer vision, and many non-negative matrix factorization (NMF)-based multiview clustering approaches have been proposed. However, most existing NMF-based multiview clustering methods aim to only push the learned latent feature matrices of all views towards a common consensus representation or only consider the consistency among all multiview data, whereas the complementary information between different views is often ignored. In this work, we propose a novel multi-view clustering via consistent and specific nonnegative matrix factorization with graph regularization method (MCCS), where the consistency among all multiview data and view-specific information in each view data are simultaneously considered. Specifically, the NMF problem is first formulated using a shared consistent basis matrix, consistent coefficient matrix, a set of view-specific basis matrices, and view-specific coefficient matrices. Then, manifold regularization is embedded into the objective function to preserve the intrinsic geometrical structure of the original data space. Furthermore, a disagreement term is designed to push these view-specific coefficient matrices further towards a common consensus and to ensure that the multiple views have the same underlying cluster structure. Moreover, the multiplicative update algorithm is employed to optimize the objective function. Extensive experimental results on five multiview benchmark datasets, namely, BBC, BBCSport, 20NGs, Wikipedia, and Handwritten, demonstrate that the proposed MCCS outperforms state-of-the-art methods, achieving improvements of 2.29%, 6.63%, 16.15%, 6.51%, and 2.85%, respectively, over the MVCC method in terms of NMI.