Abstract:Semi-supervised nonnegative matrix factorization (NMF) methods have found extensive utility in data clustering applications. However, these existing methods encounter challenges in effectively leveraging the limited supervisory information to enhance the performance of clustering. In particular, the majority of these methods tend to solely utilize either the label information or the pairwise constraint information, neglecting the potential benefits of their joint utilization. To address this issue, a novel algorithm named semi-supervised pivotal-aware nonnegative matrix factorization (SPNMF) is proposed in this paper, which utilizes the label information and the pairwise constraint information simultaneously to enhance the performance in data clustering tasks. The proposed method has two main innovations: (1) adopting the dual constraint propagation (DCP) algorithm (i.e., label propagation and pairwise constraint propagation) to effectively utilize a meager amount of label information for learning a compact data representation; (2) incorporating the pivotal-aware technique to mitigate the negative impact of outliers. Notably, the DCP algorithm not only propagates limited label information to a multitude of unlabeled samples, but also disseminates the pairwise constraint supervisory information obtained from the labels to unconstrained samples, allowing for the acquisition of richer supervisory information in the form of pointwise and pairwise constraints. Furthermore, a comprehensive analysis of SPNMF is conducted. Extensive experimental results are executed on eight real-world image datasets. The results underscore the superior performance of SPNMF in comparison to several state-of-the-art methods.

An NMF-framework for Unifying Posterior Probabilistic Clustering and Probabilistic Latent Semantic Indexing

Document Clustering Using Locality Preserving Indexing

MPC: Multi-view Probabilistic Clustering

A unified statistical approach to non-negative matrix factorization and probabilistic latent semantic indexing

Document Clustering Based on Non-Negative Matrix Factorization

Multi-View Data Fusion Oriented Clustering via Nuclear Norm Minimization

A new descriptive clustering algorithm based on Nonnegative Matrix Factorization

Document Clustering Based on Spectral Clustering and Non-negative Matrix Factorization

Semi-supervised Non-negative Patch Alignment Framework

Efficient Probabilistic Latent Semantic Analysis with Sparsity Control

From Penalized Maximum Likelihood to Cluster Analysis: a Unified Probabilistic Framework of Clustering.

Note on Algorithm Differences Between Nonnegative Matrix Factorization and Probabilistic Latent Semantic Indexing

Nonnegative Latent Factor Analysis-Incorporated and Feature-Weighted Fuzzy Double $c$-Means Clustering for Incomplete Data

Field independent probabilistic model for clustering multi-field documents

Semi-supervised pivotal-aware nonnegative matrix factorization with label and pairwise constraint propagation for data clustering

Manifold NMF with L21 Norm for Clustering

Graph Based Semi-supervised Non-negative Matrix Factorization for Document Clustering

Integrating Clustering and Multi-Document Summarization by Bi-Mixture Probabilistic Latent Semantic Analysis (PLSA) with Sentence Bases.

Multi-modal Multi-view Clustering based on Non-negative Matrix Factorization

Document clustering based on nonnegative sparse matrix factorization

Relaxed multi-view clustering in latent embedding space