Bayesian Text Classification and Summarization via a Class-Specified Topic Model

Abstract

We propose the Class Specified Topic Model (CSTM) to deal with the tasks of text classifi cation and class-specifi c text summarization. The model assumes that, besides a set of latent topics that are shared across classes, for each class there is a set of class-speci c latent topics. Each document is a probabilistic mixture of the class-specifi c topics associated with its class and the shared topics. Each class-specifi c or shared topic has its own probability distribution over a given dictionary. We develop Bayesian inference of CSTM in the semi-supervised scenario, with the supervised scenario as a special case. We analyze in detail the 20 Newsgroup dataset, a benchmark dataset for text classifi cation, and demonstrate that CSTM has better performance than a two-stage approach based on latent Dirichlet allocation (LDA), several existing supervised extensions of LDA, and a L1 penalized logistic regression. The nice performance of the CSTM is also demonstrated through Monte Carlo simulations and an analysis of the Reuters dataset