Advancing an LDA-GMM-CorEx topic model with prior domain knowledge in information systems research

Abstract

Embedding topic models with domain knowledge is deemed to be effective in bolstering the models’ interpretability. Nevertheless, contemporary topic modeling techniques introduced in past studies lack consideration for circumstances in which prior domain knowledge either does not exist or becomes obsolete quickly. Combining the latent Dirichlet allocation (LDA) with the Gaussian mixture model (GMM) and the anchor correlation explanation (CorEx) topic model, we advanced a novel LDA-GMM-CorEx topic modeling approach to enhance the domain knowledge model's adaptability and improve the interpretability of topic modeling. We further verified the effectiveness of our proposed topic modeling approach on two separate datasets from different domains, thereby attesting to its general applicability.