BERTopic

BERTopic is an open-source Python library for researchers and developers who need to find themes in document collections. It groups texts using transformer embeddings, then uses class-based TF-IDF to identify words that describe each group. Those keywords give you a way to inspect what a topic covers and which documents belong to it. The license is MIT.

You can let topics emerge from the data or guide them with seed words, predefined topics, or labeled examples. Hierarchical modeling exposes relationships between topics, while dynamic modeling tracks them over time. Incremental modeling handles documents in batches, and multilingual and multimodal approaches support analysis across languages and with images.

The modeling components are replaceable. Its default combination uses sentence-transformers, UMAP, HDBSCAN, and c-TF-IDF, but you can choose different embedding or clustering methods to suit your data. Supported embedding backends include Flair, Gensim, and spaCy.

For reviewing results, BERTopic provides document maps, topic hierarchies, keyword charts, and similarity heatmaps. You can inspect representative documents, merge overlapping topics, reduce outliers, and assign topics to new documents. It also supports several descriptions of the same topic at once, including KeyBERTInspired keywords and optional labels or summaries generated through OpenAI's cloud API with models such as gpt-4o-mini.