Rediscover your keyword search: Expand, Enrich and Rewrite
summary
Shows practical steps to get semantic ranking without full vector search by adding three concrete pieces: neural sparse (SPLADE/OpenSearch) expansion to generate token+weight rank_features stored at ingest using the OpenSearch ML connector to call models hosted on SageMaker, LLM query-to-filters rewrite to turn “black jacket for men under 120” into structured DSL filters, and image/text enrichment via AWS Rekognition/Comprehend. Benchmarks on 18 datasets show +≈5 NDCG vs standalone lexical or dense models with no fine‑tuning, ~7.9% RAM advantage over dense, and document‑only encoders keep latency competitive; demo retail app on EC2 proves improved relevance.
AI-written summary · about
chapters (9)
- 0:00Introduction to search
- 1:00Sparse vs. dense retrieval
- 6:07Text expansion techniques
- 12:55OpenSearch neural sparse
- 19:38Live search demo
- 22:21Migration and benchmarks
- 26:50Enrichment and query rewriting
- 31:05Advanced query demo
- 37:20Q&A and final takeaways
related talks
47mBoosting AI With OpenSearch: Vector Stores, Similarity Search, and Retrieval-Augme...
45mRevolutionizing Search with Power of Neural Sparse and Dense Vectors
51mQuery Understanding With Large Language Models: Techniques and...
38mNatural Language Search-Lexical and Semantic
45mMaking semantic search & RAG real: How to build a prod-ready app
30mBeyond Keywords: Building an AI-Powered Search That Reads Your Mind
38mAI-Ready Data Foundation Using Amazon OpenSearch
44mVector Search Isn’t Enough