516 Big Data Computing
in the entire corpus under consideration. The key assumptions of generative
approaches to topic modeling are
1. A topic is a probability distribution over terms. Terms can appear
in several topics. (Unique terms to a given topic are referred to as
anchor terms.)
2. A topic model on the document level is a probability distribution
over topics. Thus, a document is essentially modeled as a specic
mixture of topics.
An approach based on these considerations is latent Dirichlet analysis
(LDA), which has been implemented in several open source software dis-
tributions (Blei et al., 2003b; Heinrich, 2008). The merit of LDA is to inte-
grate a topic-based document generation model with Bayesian inference. ...