Topic Models with Meaningful Topics


Classic approaches to topic models produce clusters of documents using word co-occurrence frequencies. While these approaches are useful they don’t always align with clusters that are meaningful to humans since they don’t understand words used in different senses, underweight domain specific terminology, and don’t capture latent semantics in texts. Additionally, topic labels tend to be a bundle of words that don’t align with how humans understand topics. In this talk, we will provide an overview of contemporary approaches to topic modeling that leverage language models to produce clusters and generate meaningful topic clusters. This talk focuses on practical application and comparison of various methods to produce topics for groups of citation statements.

Aug 18, 2022 3:00 PM — 4:00 PM
QSS talks
Rowe 1014
6100 University Ave, Halifax, NS B3H 4R2