Increase text search speed & accuracy with Amazon OpenSearch Service
summary
LexisNexis builds a semantic search layer on Amazon OpenSearch Service using k-NN plugin and Google BERT embeddings, ingesting shredded passages from a 3 PB corpus (1 trillion connections) into vectors via Spark (≈24M docs reprocessed in hours). Production results: ~3M recommendations/month, near‑real‑time index updates (<5 minutes for new Supreme Court decisions), P90 latency improved most on Graviton2 R6g nodes (over 50% for batch), and human nDCG at rank 3 rises from 6.36 to 11.24 (≈75% quality gain).
AI-written summary · about
chapters (12)
- 0:00Introduction to LexisNexis
- 1:12Legal database architecture
- 2:26Why Amazon OpenSearch
- 3:40K-Nearest Neighbor plugin
- 6:30Lexis+ and research tools
- 7:36Open search architecture
- 9:52From closed to open systems
- 13:36Customer and system requirements
- 16:40Technical workflow and implementation
- 19:23Performance testing
- 22:53Results and final thoughts
- 25:20Q&A and closing
related talks
41mScaling Search and Powering Vector-Based Retrieval With OpenSea...
41mE-Commerce search at scale on Apache Lucene
19mScaling Semantic Search to Billions with Neural Sparse ANN
30mBoosting OpenSearch Performance: Lucene Bulk Collection and gRPC Search in Action
26mSimilarity search: Take your search engine to the next level
19mEnhance Search Performance At Scale with
30mModern ETL for Text Vector Data Using Apache SeaTunnel and Opensearch
28mSearch and Sushi; Freshness Counts