Generating a Billion Personal News Feeds

on YouTube ↗22m

" summary

Facebook ranks feed for 1 billion users across hundreds of billions of posts by retrieving and scoring candidates in real time, targeting under 200 milliseconds for retrieval and noting the top story gets about twice the interactions of the next item. Models predict probabilities for clicks, likes, comments and combine them with weighted sums; feature selection uses boosted trees on subsamples and negative under‑sampling, then a stacked boosted trees → logistic regression serves fast scoring, with surveys used to calibrate event weights.

AI-written summary · about

" chapters (7)

1. 0:00Introduction to news feed
2. 1:03System scale and goals
3. 4:15How the news feed works
4. 7:33Machine learning approach
5. 11:20Leveraging network data
6. 13:21Feature and model engineering
7. 17:31Measuring and balancing

" related talks

• 03mServing Billions of Personalized News Feeds with AIMeihong Wang · Artificial Intelligence Conference 2018 New York
• 32mFeed generation at ShareChatAyush Mittal · The Fifth Elephant 2019
• 54mOn How Machine Learning and Auction Theory Power Facebook AdvertisingEric Sodomka · Algorithmic Game Theory and Practice 2015
• 20mPredicting content virality to drive social trafficSriram Subramanian · Big Data & AI Toronto 2020
• 18mInstagram in focusMike Krieger · Web Summit 2015
• 30mPredictive Analytics in Social Media and Online Display AdvertisingMahesh Kumar · PASS Business Analytics Conference 2015
• 19mRushin Shah, Engineering Manager, FacebookRushin Shah · MLconf SF 2017
• 22mFacebook data and machine learning predicting elections?Alen Kisic · DSC Europe 23