Multifaceted Privacy: How to Express Your Online Persona without Revealing Your Sensitive Attributes

Recent works in social network stream analysis show that a user's online persona attributes (e.g., gender, ethnicity, political interest, location, etc.) can be accurately inferred from the topics the user writes about or engages with. Attribute and preference inferences have been widely used to serve personalized recommendations, directed ads, and to enhance the user experience in social networks. However, revealing a user's sensitive attributes could represent a privacy threat to some individuals. Microtargeting (e.g.,Cambridge Analytica scandal), surveillance, and discriminating ads are examples of threats to user privacy caused by sensitive attribute inference. In this paper, we propose Multifaceted privacy, a novel privacy model that aims to obfuscate a user's sensitive attributes while publicly preserving the user's public persona. To achieve multifaceted privacy, we build Aegis, a prototype client-centric social network stream processing system that helps preserve multifaceted privacy, and thus allowing social network users to freely express their online personas without revealing their sensitive attributes of choice. Aegis allows social network users to control which persona attributes should be publicly revealed and which ones should be kept private. For this, Aegis continuously suggests topics and hashtags to social network users to post in order to obfuscate their sensitive attributes and hence confuse content-based sensitive attribute inferences. The suggested topics are carefully chosen to preserve the user's publicly revealed persona attributes while hiding their private sensitive persona attributes. Our experiments show that adding as few as 0 to 4 obfuscation posts (depending on how revealing the original post is) successfully hides the user specified sensitive attributes without changing the user's public persona attributes.