Recommendations at scale with Elastic Search
summary
Shows a concrete pipeline using Elastic Search for content recommendations: a demo index of LEGO products, scripted scoring that combines tag scores, log-transformed social signals and exponential index-time decay, user interest vectors for personalization, plus matrix factorization / collaborative-filtering to generate tags. Measurement uses Mechanical Turk, A/B tests and benchmarks. Performance notes include tradeoffs between query vs index work and scripting choices, with a claim of 1,000 req/s, 1.2B items, <100ms if engineered carefully.
AI-written summary · about
chapters (6)
- 0:00Start
- 5:44MODEL YOUR CONTENT
- 12:28WRITE RANKING SCRIPTS
- 12:45How elasticsearch scoring works?
- 28:33Collaborative filtering KNN Matrix Factorization
- 32:43Let's talk about performance & scale
related talks
26mFrom Spark to Elasticsearch and Back
37mBeyond the basics with Elasticsearch
01h 00mElasticsearch: you know, for more than search
38mRecommendation Engine with Neo4j, Piece of Cake
41mTips and Tricks to Scale Elasticsearch for 1M RPMs and Beyond
41mBeyond the basics with Elasticsearch
38mScaling an online search engine to thousands of physical stores
46mYour own recomendation Engine with Neo4j AND Reco4PHP