Soft Subspace Clustering Ensemble Framework Based on the Latent Model

Abstract

Clustering ensemble approaches usually have more accurate, robust and stable results than traditional single clustering approaches. However, clustering ensemble can still be improved in the following aspects: (1) improve the diversity of subspaces; (2) employ probabilistic latent clustering; (3) adopt the internal latent factor analysis before the consensus function. Therefore, we propose a new clustering ensemble framework. Specifically, we analysis the original data via Jensen-Shannon divergence distribution, and then soft subspaces are generated according to the corresponding fuzzy matrix. Next, the probabilistic latent semantic analysis clustering algorithm is employed to perform clustering in each soft subspace. The final clustering performance is improved due to the reason that the latent factor model is applied to fusion matrix. Compared with traditional single and ensemble clustering algorithms, our framework achieves superior performances on 12 real-world datasets.