Text categorization based on topic model

Abstract

In the text literature, many topic models were proposed to represent documents and words as topics or latent topics in order to process text effectively and accurately. In this paper, we propose LDACLM or Latent Dirichlet Allocation Category Language Model for text categorization and estimate parameters of models by variational inference. As a variant of Latent Dirichlet Allocation Model, LDACLM regard documents of category as Language Model and use variational parameters to estimate maximum a posteriori of terms. Experiments show LDACLM model to be effective for text categorization, outperforming standard Naive Bayes and Rocchio method for text categorization. © 2008 Springer-Verlag Berlin Heidelberg.