Auto-adjustable dual-information graph regularized NMF for multiview data clustering

Abstract

Multiview data processing has gained significant attention in machine learning due to its ability to integrate complementary information from diverse data sources. Among various multiview clustering methods, non-negative matrix factorization (NMF)-based approaches have shown strong potential. However, existing methods rely on fixed, single-loss functions and single manifold regularization terms, which limit their adaptability to diverse and heterogeneous datasets. To address these challenges, we propose the multiview auto-adjustable robust dual-information graph regularized non-negative matrix factorization (MARDNMF). This method introduces a novel set of dynamically adjustable loss functions, each incorporating two correntropy terms, which are tuned via adaptive parameters based on the data characteristics. Additionally, MARDNMF leverages multi-scale k-nearest neighbors (KNNs) to build a dual-information graph regularization term, capturing both local and discriminative manifold information. Experimental results across various datasets demonstrate that MARDNMF outperforms existing NMF-based methods in both single view and multiview clustering scenarios, offering enhanced robustness and adaptability.