← News

Models & toolsSources: arXiv · 18 Nov 2026

Hierarchical Embedded Topic Model links document sections in embedding space

Researchers introduce hETM, a topic model that connects a document's topic distribution with those of its sections, reporting strong perplexity and clustering results on structured collections.

In plain words

Topic models sort documents into themes, but most treat each document as one lump of topics. The new hETM method instead links a document's themes to the themes of its individual sections, such as chapters in a legal or government report. In the authors' tests it matched or beat existing models on prediction and clustering tasks.

What was announced

Embedded topic models fold word embeddings into topic generation, which the authors say yields coherent topics and handles large vocabularies well, but they represent each document with a single topic distribution and ignore internal structure. The team proposes the Hierarchical Embedded Topic Model, which couples a document's topic distribution with those of its sections through a hierarchical prior in the embedding space. They also derive an amortized variational inference procedure to fit the model.

Evaluation covered structured collections from legal documents, company and government reports, and online discussion forums. According to the paper's abstract, hETM reached the lowest or near-lowest held-out perplexity among flat and hierarchical topic model baselines, with comparable or higher topic quality, and achieved the best clustering performance on most collections in a downstream clustering task. The reported results come from the authors' own experiments and have not been independently verified.

What it does not show

  • Evidence is limited to the paper's abstract; full experimental details, datasets and baselines are not available for verification.
  • All performance figures are the authors' own claims and do not constitute independent confirmation.

Sources

Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 19 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.