A convergent algorithm for orthogonal nonnegative matrix factorization

Abstract

This paper proposes a convergent algorithm for nonnegative matrix factorization (NMF) with orthogonality constraint on the factors. We design the algorithm based on the additive update rule algorithm for the standard NMF proposed by Lee and Seung, and derive the convergent version by generalizing the convergence proof of the algorithm developed by Lin. Further we use the proposed algorithms to improve clustering capability of the standard NMF using the Reuter document corpus, a standard dataset in clustering research.