Corinna Cortes, Head of Research, Google
summary
Google deploys Structured Snippets (launched Q3 2014) that extract facts from web tables and text, and combines them into probabilistic facts via a Knowledge Vault–style fusion. Crawls find about 11 billion tables but under 2% contain real facts; a simple subject-column detector reaches ~94% accuracy. Text extractors, graph-path features and table scores feed a combiner built with Lattice Regression that enforces monotonicity, scales to dozens of features, and pushes confidence scores toward 0 or 1.
AI-written summary · about
chapters (6)
- 0:00Fact finding at web scale
- 5:10Extracting structured data from tables
- 15:03High-quality tables on the web
- 23:53Facts from text with Knowledge Vault
- 30:22Combining signals with isotonic regression
- 36:23Summary
related talks
41mGetting Started with Query Understanding #bbuzz
38mCorinna Cortes, Head of Research, Google, NYC 2017
18mFree Texts to Knowledge Graphs - Holy Grail of Semantic Computing on an Industrial Scale
15mSemantic Search - Potential and Opportunities
27mKeynote - Google Cloud, Machine Learning
45mIntegrate your Search Engine (...)
45mFrom Buzz to Benefits: Google AI in Action
40mAnalyzing Text Data with Google Sheets and Cloud Natural Language