From text to knowledge via ML algorithms - the Quora answer
summary
Quora claims over 80 million monthly visitors and uses machine learning to turn short Q&A text into actionable signals: a personalized learning-to-rank pipeline (logistic regression, tree ensembles, LambdaMART) for feed and email digests, a hybrid moderation cues system that routes uncertain spam/harassment cases to humans, a duplicate-question binary classifier trained on merged-question labels, a topic-labeler using a large ontology, and an ask-to-answer bidirectional recommender; these systems run in production and materially improve relevance, quality and moderation throughput.
AI-written summary · about
chapters (13)
- 0:00Introduction and audience
- 1:21Quora's mission and knowledge growth
- 3:50Key dimensions: relevance, quality, demand
- 8:01Data ecosystem and network overlays
- 11:35Answer ranking and quality definition
- 17:22Moderation and spam detection
- 19:48Feed ranking and personalized relevance
- 24:43Email digest and offline ranking
- 26:07Ask to answer recommendations
- 28:14User and topic recommendations
- 30:21Understanding text relations
- 36:19Models and algorithms overview
- 37:02Conclusion and future work
related talks
25mXavier Amatriain, VP of Engineering, Quora @ MLconf SEA
42mXavier Amatriain, VP of Engineering, Quora
18mBuilding a Machine Learning Platform at
34mBuilding Literate Machines
34mIs That a Duplicate Quora Question?
31mWhen humans teach machines
22mYottaanswers.com – Delivering Billions of Smart Answers Efficiently
17mDaniel Shank, Data Scientist, Talla