Level Up: Everything you need to know about search, ...
summary
Argument: modern search moves beyond keywords by combining Lucene/Solr, Spark, Mahout, NLP and machine learning to boost relevance and cut infrastructure. Examples: Solr as KV store powers customers doing 100–200k writes/sec and even millions/sec; user behavior is ingested back into Solr to democratize signals; a learning‑to‑rank patch makes ML-based ranking practical. Surprises: core BM25 ranking often only reaches ~52% of needed relevance; only ~10% of users ever see a tuned “best” engine. Tactics that work: sessionization, recency/space boosts, streaming aggregations, and disciplined A/B testing to avoid lucky wins.
AI-written summary · about
chapters (18)
- 0:00Introduction and speaker background
- 1:06Common search pitfalls overview
- 3:54Relevance and user expectations
- 7:23Avoiding luck in search improvements
- 9:50Breaking search assumptions
- 12:42Best practices for search engines
- 15:50Leveraging time-based data
- 17:16Spatial data in search
- 19:23User intent and behavior modeling
- 23:03Alerting and proactive search
- 24:49Search vs database capabilities
- 25:52Data puzzles and record linkage
- 26:57Voice interfaces in search
- 29:02Relevance challenges and solutions
- 32:56Beyond basic keyword search
- 35:25Machine learning for ranking
- 36:49Free play in search development
- 38:55Q&A and closing remarks
related talks
56mBM25 is So Yesterday: Modern Techniques for Better Search Relevance in Solr
37mSearch and Recommendations: 3 Sides of the Same Coin #bbuzz
43m[Machine Learning] Creating more relevant search results with Learn to Rank
26mStreamlining Property Search Across Multiple Platforms
46mDiscovery & Analytics with Hadoop, Mahout & Solr
43mLearning to Rank for Reddit Search: A Project Retro
47mWhy, What, When and How of Running Diff Types of Sear...
22mMachine Learning: The Engine Behind Modern E-commerce Search