Hyperplane-based nonnegative matrix factorization with label information.

As one commonly used dimensionality reduction method, nonnegative matrix factorization (NMF), whose goal is to learn parts-based representations, has been widely studied and applied to various areas. However, classical NMF does not utilize any label information. In this paper, we propose a novel semi-supervised NMF algorithm named the hyperplane-based nonnegative matrix factorization (HNMF). This HNMF constructs a hyperplane for each cluster such that the labelled points are close to the corresponding hyperplane and far away from the other ones. Then, the discriminative abilities of representation vectors are greatly enhanced. Clustering experiments completed on five publicly available databases demonstrate the effectiveness of this proposed HNMF compared to the state-of-the-art methods. [ABSTRACT FROM AUTHOR]