ddangelov/Top2Vec

Combines Doc2Vec, BERT Sentence Transformers, or Universal Sentence Encoder embeddings with UMAP dimensionality reduction and HDBSCAN clustering to automatically discover topics without predefined counts or stop word lists. The contextual variant uses token-level embeddings to identify multiple topics per document and intra-document topic spans, exposing results through methods for topic distribution, relevance scoring, and token-level topic assignments.

3,109 stars and 5,399 monthly downloads. No commits in the last 6 months. Available on PyPI.