Testing the Homogeneity of Topic Distribution Between Documents of a Corpus

Abstract

A methodology is introduced to test the equality of topic distribution between documents of a corpus. This is achieved by using Latent Dirichlet Allocation (LDA) to estimate the topic distributions and the Kullback-Leibler divergence to measure the dissimilarity between the distributions. The testing approach combines Bayesian and frequentist statistics. Since the sampling distribution of the proposed statistics is unknown, a bootstrap test is suggested. The methodology is illustrated using scientific abstracts from the CMStatistics conference.