Self-Determined Reciprocal Recommender System with Strong Privacy Guarantees

TL;DR

This paper introduces a distributed reciprocal recommender system that ensures strong privacy through local differential privacy, allowing users to share randomized profiles in a peer-to-peer network while maintaining recommendation accuracy.

Contribution

It presents a novel peer-to-peer recommender system that guarantees local differential privacy, addressing privacy concerns in traditional centralized systems.

Findings

Recommendation accuracy remains acceptable under strong privacy guarantees.

The system effectively estimates profile similarities in a distributed setting.

Strong privacy does not significantly compromise recommendation utility.

Abstract

Recommender systems are widely used. Usually, recommender systems are based on a centralized client-server architecture. However, this approach implies drawbacks regarding the privacy of users. In this paper, we propose a distributed reciprocal recommender system with strong, self-determined privacy guarantees, i.e., local differential privacy. More precisely, users randomize their profiles locally and exchange them via a peer-to-peer network. Recommendations are then computed and ranked locally by estimating similarities between profiles. We evaluate recommendation accuracy of a job recommender system and demonstrate that our method provides acceptable utility under strong privacy requirements.