“Senator, We Run Ads”: Understanding Users Through Targeted Advertisements

Abstract

Targeted advertisements are central to the functioning of the modern web. Given their significance, many studies have attempted to understand the underlying mechanisms of the online advertising ecosystem. However, most of these works focus on identifying particular targeting behaviors through controlled web simulations. In order to paint a more comprehensive picture of targeted advertising for more realistic browsing patterns, I propose a novel approach that leverages live user data. In this thesis, I investigate what we can learn about advertisers and data flows from observing ads shown to individual users. I conduct what is (to my knowledge) the largest live-user study of targeted advertisements to date, and find that ads can viably be used to reconstruct personal identifiers, sensitive characteristics, demographic details, and user interests, despite the potential confounding variables that typically complicate such measurements.