Comparative Discourse Analysis Using Topic Models: Contrasting Perspectives on China from Reddit

Abstract

In this study, we conduct a comparative analysis of the linguistic features that differentiate two China-focused discussion communities with contrasting perspectives from Reddit. We utilize probabilistic topic modeling to represent submissions from both communities as distributions of latent patterns of word-usage. Using information theoretic measures, we conduct a series of quantitative comparisons between the language patterns of each community and identify salient features that distinguish the two communities relative to each other. We describe the rhetorical techniques and discursive frames implied by these features and how they are utilized by each community in discussions surrounding the Hong Kong protests during 2019. Additionally, we contribute a novel method for representing collections of documents that preserves interdependencies between topics at the document level.