The input of LDA is a collection of documents and a few parameters. The output is a model consisting of weights which can be normalized to probabilities. These probabilities come in two types: (a) the probability that a specific document generates a specific topic at a position, and (b) the probability that a specific topic generates a specific word from the collection vocabulary. After step (a), the document contains a list of topics (often repeated), which become words after step (b).