The Price of Fairness in Location Based Advertising

Firms use massive amounts of personal data to decide which advertisements to show to an individual, raising concerns of fairness and algorithmic bias. Previous work has proposed techniques to make machine learning more fair through awareness of the protected attributes of user data. However, these studies have either focused on specific tasks, been primarily theoretical, or have ignored the highly important domain of location-based advertising. In this work, we present an empirical analysis of the impact of fairness on advertising revenue using a real world example: location based ad personalization for users of Instagram. We empirically analyze the potential for inadvertent discrimination among gender and race in location-based systems, additionally showing the impact of location representation on fairness. Furthermore, we apply fairness techniques to analyze how revenue is affected when both individual and group fairness guarantees must hold. Though this work is a grounding for research into fairness in location-based ads, our methodology applies to more general advertising tasks. ACM Reference format: Chris Riederer and Augustin Chaintreau. 2017. The Price of Fairness in Location Based Advertising. In Proceedings of ACM RecSys conference, Como, Italy, August 2017 (FATREC’17), 5 pages. https://doi.org/10.18122/B2MD8C