BERTopic

BERTopic

BERTopic is a topic modeling technique that leverages transformer-based embeddings (BERT, Sentence-BERT, OpenAI) to create dense clusters of documents, then uses class-based TF-IDF to produce coherent, interpretable topic representations.

Key Features

• Transformer embeddings: Use any sentence-transformer, OpenAI, or Hugging Face embedding model as the backbone
• Modular design: Swap out any component — embedding, dimensionality reduction (UMAP), clustering (HDBSCAN), and vectorization
• Dynamic topics: Track how topics evolve over time with topics_over_time
• Guided modeling: Seed the model with keywords to steer topic discovery
• Zero-shot classification: Assign documents to pre-defined topics without training
• Visualization: Built-in Plotly visualizations — topic hierarchy, similarity heatmap, topic evolution
• Online learning: Incrementally update the model with new documents

Quick Start

pip install bertopic

from bertopic import BERTopic
from sklearn.datasets import fetch_20newsgroups

docs = fetch_20newsgroups(subset="all")["data"]
model = BERTopic(language="english", calculate_probabilities=True)
topics, probs = model.fit_transform(docs)

print(model.get_topic_info().head(10))
model.visualize_topics()

Add to ai-supply

npx ai-supply add bertopic-topic-modeling

Curated mirror of the open-source BERTopic (MIT). Get it from the source.