mentoring

Transformer embeddings for topic coherence

Researcher and mentor

Ongoing work on comparing transformer-based embeddings and classical topic models for interpretable analysis of large text corpora.

Topic models are often used to summarise large collections of documents, but the quality of a topic is not determined by a single score. This project looks at how embedding spaces, including transformer encoders, relate to topic coherence and to older methods such as LDA and NMF.

The work is part of a broader interest in representation learning for text. Mentoring details, including programme names, student names, and exact dates, are omitted until they are confirmed and, where relevant, approved for publication.