Machine Learning for Privacy-Preserving Ads
summary
Claim: Reddit replaces personal-data targeting with a pipeline that maps content to a knowledge graph, runs a binary disambiguation classifier on ambiguous tokens (example: "switch"), and embeds graph entities to compute similarity between user context and ad content. The implementation averages entity vectors, measures distances to rank ads, and surfaces higher-click-probability ads while keeping PII out of the loop; cited stats include 8/10 Americans worried about tracking and 9/10 in Japan, and the system reports improved relevance and privacy-preserving revenue stability.
AI-written summary · about
chapters (10)
- 0:00Privacy-preserving ads intro
- 1:20User fears targeted ads
- 3:07Bad ad practices problems
- 5:15Contextual ads replace data?
- 8:06Symbolic AI for context
- 11:49User context with knowledge graphs
- 13:57Ads matching with embeddings
- 16:06Advertiser platform applications
- 17:54Contextual ads benefits
- 18:38Adopt contextual standards
related talks
33mMachine Learning to moderate ads
13mAlgorithmic Transparency in Personalized Advertising: Learnings from Building "Why Am I
32mPlatform-supported Auditing Of Social Media Algorithms For Public Interest
33mMachine Learning in the age of increasing Data Privacy Consciousness
33mPractical Data Privacy
22mCommunicating Algorithmic Process in Online Behavioral Advertising
27mThe ethical algorithm
28mAdvertising, algorithms and Privacy: recent opensource developments