Assessing the Reliability of Facebook User Profiling

User profiling is an essential component of most modern online services offered upon user registration. Profiling typically involves the tracking and processing of users' online traces (e.g., page views/clicks) with the goal of inferring attributes of interest for them. The primary motivation behind profiling is to improve the effectiveness of advertising by targeting users with appropriately selected ads based on their profile attributes, e.g., interests, demographics, etc. Yet, there has been an increasing number of cases, where the advertising content users are exposed to is either irrelevant or not possible to explain based on their online activities. More disturbingly, automatically inferred user attributes are often used to make real -world decisions (e.g., job candidate selection) without the knowledge of users. We argue that many of these errors are inherent in the underlying user profiling process. To this end, we attempt to quantify the extent of such errors, focusing on a dataset of Facebook users and their likes, and conclude that profiling -based targeting is highly unreliable for a sizeable subset of users.