The 5th AAAI Workshop on Privacy-Preserving Artificial Intelligence recently accepted a paper on personalized recommendations in today’s internet landscape. Personalized recommendations are crucial for artists and creators to connect with interested users and for users to discover new and exciting content. However, many users are wary of platforms that offer personalized recommendations due to concerns about data privacy. As a result, businesses are now focusing on creating privacy-first systems.
In this article, an algorithm for personalized recommendations that ensures precise and differentially-private measurement is proposed. The algorithm is particularly useful for applications such as advertising. Offline experiments were conducted to measure how this privacy-preserving algorithm impacts key metrics related to user experience, advertiser value, and platform revenue compared to non-personalized and non-private personalized systems.