UROP Research Mentor Project Submission Portal: Submission #929

Overall Research Project Description Information dissemination is primarily controlled through platforms across their social media network infrastructure. At the heart of this system is the platform itself, which has access to a vast array of heterogeneous content. On the periphery, individual users act as information consumers. It has become widely recognized that these platforms utilize a mechanism popularly known as the "Algorithm''. The "Algorithm'' selects which information to present to users based on machine learning model (or an estimate) of content preferences specific to each consumer. The user's objective is straightforward – to maximize its utility – while the platform's aim is to maximize consumer engagement. The misalignment between the user's and platform's objectives results in intriguing behaviors that are observed empirically. In this project we propose a new game-theoretic model to analyze how content distribution platforms optimize user engagement for users interested in keeping their preferences with provable guarantees using differential privacy techniques. The contribution of the project is twofold:

1. To inform platforms on how to improve their machine learning recommendation algorithms for a user population that has strict privacy requirements;

2. To inform users on how to judiciously disclose their preferences to ensure they receive good recommendations from the platform while guaranteeing a base level of privacy protection.