Applying Topic Modeling with Prior Domain-Knowledge in Information Systems Research

Authors

Yuting Jiang, Department of Information and Service Economy, Aalto University School of BusinessFollow
Mengyao Fu, The Laboratory for AI-Powered Financial Technologies LimitedFollow
Jie Fang, University of Nottingham Ningbo ChinaFollow
Matti Rossi, Information and Service EconomyFollow

Abstract

Topic modeling is gaining traction in Information Systems (IS) research as more textual material becomes accessible online and computational tools for analyzing large textual datasets are getting more powerful. This paper advances a new two-step correlation explanation topic modeling (Corex) method with prior domain knowledge to improve the interpretability of topic modeling to meet the needs of current IS research. The proposed method combines the traditional Latent Dirichlet Allocation topic model and the Anchored correlation explanation topic model. In the first step, the approach allows for the rapid and maximum acquisition of topic words related to domain knowledge. These anchor words are then inputted into the second-step CorEx topic model. We further applied and verified the effectiveness of the two-step Corex method to a textual dataset containing 4,290,484 users’ personal profiles, thereby illustrating the utility of applying this innovative topic-modeling method in information systems research.

Comments

Paper Number 1582; Track Philosophy; Complete Paper

Recommended Citation

Jiang, Yuting; Fu, Mengyao; Fang, Jie; and Rossi, Matti, "Applying Topic Modeling with Prior Domain-Knowledge in Information Systems Research" (2023). PACIS 2023 Proceedings. 143.
https://aisel.aisnet.org/pacis2023/143