Predictive Analytics in Social Media and Online Display Advertising
summary
Uses logistic regression on hundreds of millions of impressions with ~5,000 features, retaining all clicks and sampling 1% of zeros to handle 0.1% CTR sparsity; a lift chart shows the top 20% of impressions yield 56% of clicks (≈180% lift) and the top 3–4% yield >20% of clicks. Builds a Facebook optimizer combining priors, weighted recent data and linear models across ~10,000 microsegments; baseline CTR is 0.03% and application rate ~1%, and the optimizer raises CTR by 50% and cuts cost per application by 25%.
AI-written summary · about
chapters (10)
- 0:00Predictive analytics in advertising
- 1:24Real-time bidding
- 8:08Click-through rate with regression
- 11:59Programmatic media trends
- 13:47Credit card acquisition via Facebook
- 18:22Facebook ad conversion challenges
- 22:13Micro-segmentation for Facebook
- 26:02Ad performance optimizer tool
- 27:49Results and use cases
- 29:43Q&A
related talks
54mOn How Machine Learning and Auction Theory Power Facebook Advertising
27mLarge Scale Response Prediction to Serve Ads on LinkedIn
31mLarge Scale CTR Prediction Lessons Learned
23mThe unreasonable effectiveness of data for better monetisation
38mHow to Enhance Customer Targeting in Marketing
40mNext Best Action: Maximizing Customer Value
46mFrom correlation to causation: measuring ad effectiveness at scale
25mClaudia Perlich, Chief Scientist, Dstillery, NYC 2017