Editorial illustration for LLM embeddings and HDBSCAN cluster text; visualized with pairwise scatterplots
LLM embeddings and HDBSCAN cluster text; visualized with...
LLM embeddings and HDBSCAN cluster text; visualized with pairwise scatterplots
The current wave of Generative AI often highlights chat bots and prompt engineering, but LLMs can do more. While the hype leans on conversation, one of the models’ quieter strengths is turning raw, messy text into embeddings—dense vectors that capture meaning. Here’s the thing: once you have those vectors, you can feed them into classic machine‑learning tools.
In this guide we stitch together a pre‑trained sentence‑transformers model, UMAP for dimensionality reduction, and HDBSCAN, a density‑based clustering algorithm, to surface hidden topics in an unlabeled corpus. The pipeline starts with a freely available text dataset, runs it through an open‑source embedding model, trims the dimensionality, then lets HDBSCAN discover clusters without any prior labels. It’s a step‑by‑step walk‑through, from raw documents to visual pairwise scatterplots that let you see the groupings.
By the end you’ll have a reproducible workflow that shows how embeddings and density‑based clustering can turn chaos into structure—no manual tagging required.
For extra insight, we can show some cluster visualizations with the aid of the supplementary code provided below, which shows a scatterplot for every pairwise combination of the five existing components that describe each data point: Result: By trying different configurations for HDBSCAN, you may come across results in which the number of identified clusters could be different from two. Wrapping Up Once we have gone through the process of building the text-based clustering pipeline, it is worth concluding by pointing out the key reasons why putting together LLM embeddings with HDBSCAN is worth it.
Why this matters
We’ve seen a straightforward pipeline that turns raw documents into clustered topics without any labeled data, leveraging a pre‑trained sentence‑transformers model for embeddings, UMAP for dimensionality reduction, and HDBSCAN for density‑based grouping. The article walks through each step, showing that even a modest five‑component representation can be visualized through pairwise scatterplots, giving developers a tangible way to inspect cluster structure. Yet the narrative stops short of quantifying accuracy or scalability; it’s unclear whether the approach holds up on larger corpora or more nuanced vocabularies.
The code snippets promise “extra insight,” but the usefulness of those visualizations depends on the analyst’s ability to interpret dense plots. For founders eyeing rapid topic discovery, the method offers a low‑friction entry point, though we should be cautious about assuming it replaces more elaborate supervision. Researchers may appreciate the flexibility of tweaking HDBSCAN configurations, but the article leaves open how sensitive results are to those choices.
In short, the technique is a practical addition to the toolbox, but its broader impact remains to be validated.
Further Reading
- Clustering Unstructured Text with LLM Embeddings and HDBSCAN - Machine Learning Mastery
- From Text to Insights: Hands-on Text Clustering and Topic Modeling with LLMs - OpenAI Blog
- Clustering Documents with OpenAI embeddings, HDBSCAN and LangChain - Dylan Castillo
- Large Language Model Enhanced Clustering for News Event Detection - arXiv
- LLM Text Clustering and Topic Modeling: HDBSCAN and BERTopic - Ranjan Kumar
Related Reading
LLMs & Generative AI LWiAI Podcast #228: OpenAI unveils GPT-5.2, Runway rolls out first world model
LLMs & Generative AI OpenAI's Codex powers Lovable AI, letting millions create apps from text
LLMs & Generative AI Google releases FunctionGemma, a tiny model for natural-language mobile control
From Archives AI Agents Risk Fatal Traps When Treating Context Windows as Memory
From Archives