A Survey of Topic Model Inference Techniques

Abstract

— Latent Dirichlet Allocation (LDA) is a probabilistic topic model that aims at organizing, visualizing, summarizing, searching, predicting and understanding the content of any given text data. The model enables users to discover themes in text, annotate, organize and summarize documents. LDA inference involves estimating the parameters and posterior distribution of a formulated mathematical relationship. This paper investigates topic modeling literature based on LDA and presents discoveries and state of the art in the topic. Presented also are challenges and popular tools. In conclusion, the paper identifies Gibbs sampling as a popular inference mechanism and notes that the method is limited for application in big data settings.