A generative probabilistic model that represents documents as mixtures of topics, where topics are distributions over vocabulary words.