A Comparative Empirical Evaluation of Topic Modeling Techniques

Abstract

AbstractThis paper presents empirical evaluation of three most important topic modeling techniques—latent semantic analysis, latent Dirichlet allocation and correlated topic modeling, The novelty of work exists in application or domain-independent comparative evaluation of techniques, that is not presented before in the literature. Most of the work on topic modeling is very application specific, so it becomes difficult to conclude a generalization solution. In this research, a detailed comparative evaluation of topic modeling techniques is presented, and it is found that latent Dirichlet allocation technique is the true synonym for topic modeling. So latent Dirichlet allocation is a truly unsupervised kind of machine learning technique, correlated topic modeling is another Dirichlet free probabilistic topic modelling to find correlation among topics. Latent semantic analysis presents semantic gist of corpus at the diagonal in such an effective way, that it can be considered as a very promising technique in natural language processing and text mining.KeywordsLatent semantic analysisLatent Dirichlet allocationBag of Words (BoW)Dimension reduction