Xavier Amatriain, VP of Engineering, Quora @ MLconf SEA
summary
Quora builds concrete ML systems: an answer-ranking using learning-to-rank with LambdaMART that emphasizes text quality and user expertise (currently unpersonalized), a personalized multi-stage feed ranker, duplicate-question binary classifiers using vector-space and usage features, and a real-time A2A real-time answerer suggestions model that ranks likely answerers as a question is typed. Spam uses hybrid moderation cues (ML scores queue human reviewers). Models include logistic/elastic net, GBDT, random forests, neural nets, matrix factorization and LDA; company size ~110, engineering ~75.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:10Quoras mission
- 3:56Complex network
- 4:31Semantics
- 4:57What do Quora care about
- 6:15Quoras machine learning applications
- 6:36Answer ranking
- 7:59Features
- 11:43Recommendations
- 12:43Related Questions
- 14:05Duplicate Questions
- 15:20User Trust and Expertise
- 16:58Trending Topics
- 17:58Spam Detection
- 19:28Content Creation Prediction
- 21:26Models
- 23:11Conclusions
related talks
42mXavier Amatriain, VP of Engineering, Quora
38mFrom text to knowledge via ML algorithms - the Quora answer
04mXavier Amatriain interview
18mBuilding a Machine Learning Platform at
23mLive Q&A
17mDaniel Shank, Data Scientist, Talla
34mBuilding Literate Machines
26mThe geometry of Intelligence in a Large Question-Answering System