Industrial Applications of Machine Learning in Search
summary
Microsoft Newsfeed content understanding builds hierarchical content signals from title, snippet and multimedia using a pre-trained Transformer multitask model over a 100K taxonomy, links keyphrases to a millions-scale Wikipedia-backed entity repo, and achieves entity extraction 93% precision. Ongoing platform work adds self-serve model training, dashboards and measurement sets to handle multilingual coverage, taxonomy drift and labeling guidelines. A topic propensity model without collaborative filtering yields a measured 1.25% engagement lift, default topic recall gives 0.1–2% lift.
AI-written summary · about
chapters (5)
- 0:00Content understanding in feeds
- 6:26Signals from content team
- 18:43Existing models and challenges
- 27:43Business impact and onboarding
- 33:54Ongoing work and future
related talks
42mThe Machine Understands Text — Learning and Search at Micro and Macro Scale
24mMachine Intelligence gets real
22mMachine Learning: The Engine Behind Modern E-commerce Search
51mTaking Models to the Next Level with Azure Machine Learning
30mSemantic Search and Neural Information Retrieval
31mWhen humans teach machines
41mResearch in the Age of AI
58mExploring Machine Learning for .Net Developers