Herding Entities- Information Search and Synthesis in the Context of Transaction Data
summary
Ntropy claims a production entity search that enriches noisy transaction text by combining Elasticsearch indexes, milvos vector nearest‑neighbor lookups, and a cross‑encoder re‑ranker from sentence‑transformers, with caching and fuzzy caching to cut API cost. The pipeline groups duplicate URLs, boosts results with label data and domain metadata, then enriches with logos, locations and roles. The system ships real‑time and batch modes, and the end‑to‑end loop uses human labeling to steadily raise precision.
AI-written summary · about
chapters (12)
- 0:00Transaction intelligence intro
- 1:53Transaction enrichment at Entropy
- 3:23Entity search challenges
- 7:02Entity search approach
- 11:55External vs internal search engines
- 17:15Caching for cost and time
- 19:04Internal entity search algorithm
- 24:03Re-ranking entity results
- 28:19Finalizing entity enrichment
- 31:31System architecture and modes
- 32:34Feedback loops for improvement
- 33:39Q&A
related talks
49mNamed-Entity Recognition and Name Resolution Using Similarity Queries
44mNeedle in a haystack : entity search on text and graph
30mSemantic Search and Neural Information Retrieval
30mSearch on Steroids — A Story of Semantics
36mUnderstanding your Bank Statement in 100ms
33mA Practical Approach To Semantic Search
22mHarnessing image and text data for product matching
22mMachine Learning: The Engine Behind Modern E-commerce Search