Abstract

Advertising is the primary source of revenue for many mobile apps. One important goal of the ad delivery process is targeting users, based on criteria like users’ geolocation, context, demographics, long-term behavior, etc. In this paper we report an in-depth study that broadly characterizes what targeting information mobile apps send to ad networks and how effectively, if at all, ad networks utilize the information for targeting users. Our study is based on a novel tool, called MAdScope, that can (1) quickly harvest ads from a large collection of apps, (2) systematically probe an ad network to characterize its targeting mechanism, and (3) emulate user profiles of specific preferences and interests to study behavioral targeting. Our analysis of 500K ad requests from 150K Android apps and 101 ad networks indicates that apps do not yet exploit the full potential of targeting: even though ad controls provide APIs to send a lot of information to ad networks, much key targeting information is optional and is often not provided by app developers. We also use MAdScope to systematically probe top 10 in-app ad networks to harvest over 1 million ads and find that while targeting is used by many of the top networks, there remain many instances where targeting information or behavioral profile does not have a statistically significant impact on how ads are chosen. We also contrast our findings with a recent study of targeted in-browser ads.