Performance Evaluation of GNMF-based Clustering Using Cluster Ensembles

Abstract

In recent years, Graph regularized Nonnegative Matrix Factorization (GNMF)-based clustering has attracted attention as a promising clustering method for large-scale high-dimensional data. GNMF was developed by applying the idea of manifold learning to NMF, and thus can conduct dimensionality reduction suitable for real-world data. However, minimizing the objective function for GNMF does not always result in a high clustering performance, because there is a gap between the objective function value and the clustering performance. In addition, there is a problem that the clustering results depend on the initial value, which has not been discussed in detail in many papers. In this paper, as a method to solve these problems, we propose a GNMF-based clustering using cluster ensembles. We also show experimentally that the proposed method stably achieves a high clustering performance.