This research project evaluates the classic Latent Dirichlet (LDA) and neural topic modeling with the application to a novel corpus of research in the domain of Education, and further investigates the "topic collapsing" problem in Dieng et al. (2020)’s implementation of the embedded topic model (ETM) . We highlighted aspects of the ETM’s behavior that suggest drawbacks of topic modeling in embedding spaces, as well as a potential solution to the topic collapsing problem.
- AUTHORs: Jiner Zheng, Jon Ball