Figure 2

Informal explanation of the intuition behind topic modelling (adapted from (Blei[2012])). An excerpt of a sample abstract (Butt et al. [2013]) is quoted for illustration purposes. A document is assumed to be the result of an iterative process that selects topics from a document-specific probabilistic topic distribution and words from a topic-specific probabilistic word distribution. Topic modelling algorithms reverse this process in order to find assignments for the two distributions that best explain a set of documents.