Concepedia

Probabilistic Topic Models

David M. Blei, Lawrence Carin, David B. Dunson

IEEE Signal Processing Magazine · 2010 · 339 citations · 54 references

Concepts

ComputingComputer ScienceData ScienceData MiningInformation RetrievalText AnalyticsCorpus AnalysisDocument RepresentationText MiningText RepresentationTopic ModelingUnsupervised LearningProbabilistic Topic Models For RetrievalUnsupervised Topic Modeling

TL;DR

Topics are distributions that capture salient themes across a document collection. The article reviews probabilistic topic models, graphical models that summarize large document collections with fewer word distributions. The review covers finite parametric topic models, Bayesian nonparametric HDP models, and extensions for time‑varying topics and topic prevalence. The authors demonstrate topic models applied to nontext data, highlighting recent image‑analysis research.

Abstract

In this article, we review probabilistic topic models: graphical models that can be used to summarize a large collection of documents with a smaller number of distributions over words. Those distributions are called "topics" because, when fit to data, they capture the salient themes that run through the collection. We describe both finite-dimensional parametric topic models and their Bayesian nonparametric counterparts, which are based on the hierarchical Dirichlet process (HDP). We discuss two extensions of topic models to time-series data-one that lets the topics slowly change over time and one that lets the assumed prevalence of the topics change. Finally, we illustrate the application of topic models to nontext data, summarizing some recent research results in image analysis.