Multi-View Clustering of Dual Graph Non-Negative Matrices Factorization with Diversity Constraints

Non-negative matrix factorization (NMF) has become a hot-spot of extensive research due to its ability to achieve dimensional reduction, but the existing research still has limitations: 1) they only utilize information about the consistency of the data while ignore information about the diversity due to the multifaceted collection of data. 2) They capture only the local structure of the data, while useful information that can facilitate the clustering task is still present in the feature manifold. To alleviate the above problems, we propose multi-view clustering of dual graph non-negative matrices factorization with diversity constraints (MVC-DGNMF). Specifically, diversity constraints are utilized to capture diversity information between views. Then, the data and feature manifold are utilized to extract internal structural information, taking diversity into account. Finally, an algorithm for efficient iterative optimization is solved based on the model in this paper. Experiments demonstrate that the approach in this paper attains better performance compared to the excellent multi-view clustering methods.