Search and Recommendations: 3 Sides of the Same Coin #bbuzz
summary
Compares three concrete ways to serve recommendations with search: score then search (separate scorer and searcher, flexible models but round‑trip latency), native search (precompute co‑occurrence or thresholded indexes for sub‑50ms serving at scale) and custom ranking (Elasticsearch vector scoring plugin / Solr port doing online dot‑product scoring with payloads). Shows locality‑sensitive hashing and LSH precomputation tradeoffs, gives ~50ms targets, and cites a claimed up to 10x plugin speedup while noting dimensionality and shard tradeoffs.
AI-written summary · about
chapters (10)
- 0:00Introduction and background
- 1:10Recommender systems overview
- 9:16Core recommendation models
- 12:28Search engines for recommendations
- 14:35Three integration approaches
- 17:46Native search approach
- 22:25Custom ranking approach
- 29:11Performance evaluation
- 34:30Summary and conclusion
- 35:34Q&A session
related talks
40mLevel Up: Everything you need to know about search, ...
38mLearning to Rank: where search meets machine learning #bbuzz
47mWhy, What, When and How of Running Diff Types of Sear...
01h 14mBuilding a Recommendation System with a Hybrid Search From Scratch
32mLeveraging recommender systems to personalise search results
01h 09mVector Search: Ask Me Anything!
23mML-Powered Search & Recommendation 101
21mRecsys Keynote: Improving Recommendation Systems & Search in the Age of LLMs