BERTopic
A Python library for topic modeling that combines BERT-style embeddings with class-based TF-IDF to produce dense, easily interpreted topic clusters.
BERTopic is a topic-modeling library that uses transformer embeddings together with a class-based variant of TF-IDF to build dense clusters of documents, keeping the words that matter in each topic's description so the resulting topics stay easy to interpret.
It supports guided, supervised and semi-supervised topic modeling; manual, multi-topic and hierarchical variants; class-based, dynamic and online/incremental modeling; and multimodal and multi-aspect modeling, alongside newer zero-shot topic modeling, merging separately trained models and seeding topics with chosen words.
It installs with pip or uv, optionally with alternative embedding backends such as Flair, Gensim, spaCy or Universal Sentence Encoder, or a lighter installation without transformers, UMAP or HDBSCAN for use with other embedding methods.