Beyond Opt-Outs: Quantifying User Notification Harm
summary
Presents a measurable framework that predicts notification harm by combining feature engineering, ML models (Lasso and Random Forest), free‑text clustering with LLMs and user surveys (Google Opinion Rewards and Cint). Key finding: fast dismiss behavior—especially time-to-dismiss <30s—correlates with later opt‑outs and lower satisfaction; dismissal rates above ~60% reverse expected signals. Teams instrument the new signal in A/B models across Google Photos, Maps, Search and YouTube, and report it surfaces dissatisfaction earlier so product rules change accordingly.
AI-written summary · about
chapters (6)
- 0:00Introduction to talk
- 6:32Research journey begins
- 11:18Key findings from data
- 15:39User insights reveal patterns
- 21:45AI analysis of user feedback
- 22:50Why this matters for teams
related talks
25mIntelligent Notification Systems at Scale
27mMaking Money with Data Science
21mRise of the Machines: Removing the Human-in-the-Loop
42mOnline Safety Bill: How Global Platforms Use MLOps to Keep People Safe
13mClaudia Perlich, Chief Scientist, Dstillery
12mPredictive Response Optimization: Using Reinforcement Learning to...
18mEvaluating the Disruptiveness of Mobile Interactions: A Mixed-Method Approach
36mFraud in mobile applications: how to define and detect