Hybrid Search with Apache Solr
summary
Apache Solr supports vector KNN since 9.0 via the KNN Query Parser and adds prefiltering behavior in 9.1 so filters apply during top-K construction (avoiding empty result sets); 9.6 introduces local params for prefilter/include/exclude tags to control which fq are applied early. Retrieval options include union or intersection of lexical and vector candidate sets; ranking uses normalized score functions, product, or a trained model — Solr’s Learning to Rank vector feature (since 9.3) combines lexical and vector similarity as features but optimal weighting remains an open problem.
AI-written summary · about
chapters (8)
- 0:00Introduction and speaker
- 1:58Limitations of vector search
- 8:06What is hybrid search
- 9:35Hybrid search in Solr
- 13:15Prefiltering vs postfiltering
- 18:16Ranking hybrid results
- 23:23Learning to rank approach
- 25:53Future work and recap
related talks
47mNeural Search Comes to Apache Solr: Approximate Nearest Neighbor, BERT & more
41mApache Solr 10: What's Coming up for Vector Search
37mLexical, Vector & Hybrid Search with Elasticsearch
01h 09mVector Search: Ask Me Anything!
40mLevel Up: Everything you need to know about search, ...
44mVector Search Beyond the Hype: Real-World Implementation and Performance Insights
41mText to Context: How We Introduced Hybrid Search
20mUsing Apache OpenNLP with OpenSearch K-NN Vector Search