An efficient distributed privacy-preserving recommendation system
Frederik Armknecht, Thorsten Strufe · 2011
Implementing a recommendation system on the data of mobile social networks exploits knowledge about behavior and preferences of its users and hence raises serious privacy concerns. Leveraging the wealth of aggregated information in these services promises an immense benefit by allowing suggestions for presumably appreciated, yet previously unseen restaurants, sights, and further types of locations. Privacy preserving recommenders based on homomorphic encryption have been proposed, which have a systematic draw-back: while recommender systems often store their information as real values, all homomorphic encryption schemes used today process only data from other algebraic structures, e.g., the ring of integers modulo some integer n. Therefore, we present a novel distributed recommender and a homomorphic encryption scheme, which works directly on real numbers and which possesses some remarkable properties: it is conceptually simple, efficient, and provably secure.