BERTopic

Why we included this project

Reading through a large corpus and trying to spot the main themes is tedious, and BERTopic is a solid way to skip most of that work. It embeds documents with a transformer model, clusters those embeddings, and then labels each cluster with the words that most clearly set it apart, so you get readable topic descriptions instead of meaningless cluster numbers. The pipeline is modular, which matters in practice: you can swap in your own embeddings, pick a different clustering step, or plug an LLM in at the labeling stage to get richer topic names and summaries. You also have a lot of control over the kind of analysis you want, from guided or supervised runs to hierarchical, dynamic, and online variants that track how topics shift over time. For teams doing text exploration, organizing content, or watching how discussion topics trend, it is a practical starting point.