I like this thread "A parellel between LSA and pLSA" https://stats.stackexchange.com/questions/40723/a-parellel-between-lsa-and-plsa https://stats.stackexchange

I like this thread "A parellel between LSA and pLSA" https://stats.stackexchange.com/questions/40723/a-parellel-between-lsa-and-plsa https://stats.stackexchange.com/questions/40723/a-parellel-b... showing differences between PCA matrix factorization, some naive non-negative matrix factorization (with L2 cost function) and pLSA (log-loss cost function). The later is a precursor to Latent Dirichlet Allocation, one of the key models for topic modeling.