Communicating Algorithmic Process in Online Behavioral Advertising
summary
Study runs a lab with 32 participants who see real personal ads, an ad explanation view, an algorithm view (Facebook-inferred interests including odd items) and an advertiser view (Facebook advertiser tool with ~29,000 features) plus two speculative design tasks. Participants prefer specific, interpretable, non-creepy explanations that link to identity; revealing inferred attributes produces “algorithm disillusionment” as users discover noisy or wrong labels; recommendations include embedding explanations in ads and phrasing reasons abstractly rather than full transparency.
AI-written summary · about
chapters (11)
- 0:00Introduction to talk
- 0:54Study on algorithmic transparency
- 3:00Ad explanation view phase
- 4:01Speculative design test one
- 4:42Algorithm view phase
- 5:23Advertiser view phase
- 6:27User evaluation of explanations
- 11:06Hidden algorithmic variables
- 13:38Improving transparency mechanisms
- 15:54Broader algorithmic literacy
- 17:03Q&A begins
related talks
13mAlgorithmic Transparency in Personalized Advertising: Learnings from Building "Why Am I
32mPlatform-supported Auditing Of Social Media Algorithms For Public Interest
54mOn How Machine Learning and Auction Theory Power Facebook Advertising
19mMachine Learning for Privacy-Preserving Ads
43mKeynote: Consequences of an Insightful Algorithm
06m(M)ad to See Me? Intelligent Advertisement Placement: Balancing User Annoyance and Advertising ...
25mClaudia Perlich, Chief Scientist, Dstillery, NYC 2017
29mAI and ML for Computational Advertising