Why am I seeing this: Democratizing End User Auditing for Online Content Recommendations
summary
An interactive browser sandbox generates synthetic personas and substitutes their data (IP, browsing history, location, ad IDs) to test whether specific attributes change ad delivery, captures screenshots, and uses an LLM-based ad alignment score (0–100) to quantify ad-persona matches. A technical evaluation shows high ad-detection accuracy and stable scores; expert review rates persona realism; a 15-participant user study finds age and income drive theme and visual shifts and that the tool moves vague concern into actionable evidence.
AI-written summary · about
chapters (10)
- 0:00Introduction and motivation
- 1:24Privacy auditing challenges
- 2:45Privacy auditing sandbox concept
- 3:49Persona variation method
- 4:33Hypothesis consolidation process
- 5:59User case study with Alice
- 7:24Technical evaluation results
- 8:26User study findings
- 9:51Summary and takeaways
- 10:13Q&A session
related talks
06mWho Am I? A Design Probe Exploring Real-Time Transparency About Online and Offline User ...
32mPlatform-supported Auditing Of Social Media Algorithms For Public Interest
27mTracing Information Flows Between Ad Exchanges Using Retargeted Ads
25mFiltering bad ads using LLM assisted adblocking
05mDeciding on Personalized Ads: Nudging Developers About User Privacy
01h 00mWelcome to the machine
22mExploring User Perceptions of Discrimination in Online Targeted Advertising
01h 07mThe great data race ...