Getting Started with Query Understanding #bbuzz
summary
Argues that focusing on intent instead of raw keywords prevents “poisoned” results and lists concrete fixes: add entity exceptions to stop aggressive stemming, detect phrases with PMI phrase detection or Wikipedia matches, classify queries into broad buckets (finished goods vs craft supplies), harvest synonyms from WordNet/Wikipedia and human review, extract image color with Color Thief or regex tags, and combine human curation with ML (word2vec, CNNs, LTR). Notes Google’s knowledge graph ~70 billion facts as an impractical baseline for most teams.
AI-written summary · about
chapters (12)
- 0:00Introduction and speaker background
- 1:32Why query understanding matters
- 9:45Query understanding vs alternatives
- 13:15Precision problems in search
- 17:31Handling phrases in queries
- 20:24Query classification for ambiguity
- 24:59Recall problems in search
- 27:49Synonyms and missing mappings
- 30:18Human vs automated solutions
- 33:28Confidence spectrum in query handling
- 33:49Query understanding and learning to rank
- 39:05Q&A session
related talks
21mBeyond Significant Terms #bbuzz
35mYou've got questions. We've got answers! #bbuzz
40mLevel Up: Everything you need to know about search, ...
37mCorinna Cortes, Head of Research, Google
39mHow to Avoid structural biases in evaluating your Machine Learning/NLP projects
44mBuilding Knowledgeable Machines
21m"What's in the news?" #bbuzz
01h 09mVector Search: Ask Me Anything!