Topic Modeling in Finance: A Review of Methods, Applications, and Challenges

Topic modeling is one of the most widely used Natural Language Processing models in business fields. In this survey, by collecting and reviewing 140 topic modeling-related articles published in 40 finance and related business journals, I document the trend of topic modeling across journals and time, review the main algorithms used in the literature, and organize the evidence by research areas, research methodologies, and data sources. The survey shows that Latent Dirichlet Allocation is the dominant approach especially in early studies, but newer variants, such as supervised LDA, correlated topic modeling, sentence-level models, and structural topic models, are being adopted when researchers need better model performances under specific cases. Recent work increasingly uses topic-based methods to summarize documents, construct new measures, classify disclosures, and compare text information from firms, market participants, and policymakers. Though topic modeling algorithms are powerful, challenges such as noisy documents, topic labeling, and Blackbox issues still exist. Overall, topic modeling has moved from a supplementary textual analysis tool to a main research tool in finance research, and topic modeling will accelerate the development of finance research in the near future.