Data Mapping for Data Exploration
summary
Demonstrates converting embeddings into an interactive 2D map using UMAP, fast nearest-neighbor search against vector databases, and hierarchical clustering via HDBSCAN, packaged into an open-source app called TNT. Uses the CORD‑19 corpus downsampled to ~100,000 papers, attaches keyword and citation-based labels, adds semantic search and LLM-generated cluster summaries, and surfaces surprising results (distinct Computer Science and environmental clusters among COVID papers). The pipeline is implemented, public on GitHub, and the demo works end-to-end.
AI-written summary · about
chapters (9)
- 0:00Data mapping intro
- 1:03Three modes of analysis
- 2:48Query to search transition
- 8:27Why interactive maps work
- 12:47Pipeline for interactive maps
- 16:39Open-source mapping tools
- 17:02COVID-19 case study
- 23:49Adding interactivity with TNT
- 29:29Key takeaways
related talks
48mMAP All the Things
25mThe Art of Clustering: The Good, The Bad and The Beautiful
57mVisualization Tools for Data Discovery
57mDiscover semantic topics in a collection of documents
01h 00mData, visualization, and designing AI
31mBeyond Vectors: Evolving GenAI through Transformative Tools and Methods
45mWords in Space: A Visual Exploration of Distance, Documents, and Distributions...
01h 46mVector Databases (for GenAI & beyond)