Online matrix factorization for space embedding multilabel annotation

Abstract

The paper presents an online matrix factorization algorithm for multilabel learning. This method addresses the multi-label annotation problem finding a joint embedding that represents both instances and labels in a common latent space. An important characteristic of the novel method is its scalability, which is a consequence of its formulation as an online learning algorithm. The method was systematically evaluated in different standard datasets and compared against state-of-the-art space embedding multi-label learning algorithms showing competitive results. © Springer-Verlag 2013.