2005/029 - Relation between PLSA and NMF and implications
Eric Gaussier, Cyril Goutte
The 28th annual International ACM SIGIR, Conference on Research and Development in information retrieval, Salvador, Brazil, August 15-19, 2005.
The techniques of non-negative matrix factorisation (NMF,  and Probabilistic latent semantic analysis (PLSA, ) have been succesfully applied to a number of text analysis tasks such as document clustering. Despite their different inspirations, these methods are both instances of multinomial PCA . We further explore this relationship and first show that PLSA solves the problem of NMF with KL divergence, and then explore the implications of this relationship