Robust Privacy-Preserving Federated Item Ranking in Online Marketplaces: Exploiting Platform Reputation for Effective Aggregation

Online marketplaces often collect products to sell from several other platforms (and sellers) and produce a unique ranking/score of these products to users. Keeping as private the user preferences provided in each (individual) platform is a need and a challenge at the same time. We are currently used to rating items in the marketplace itself which, in turn, can produce more effective rankings. Hence, the shaping of an effective item ranking would require a sharing of the user ratings between the individual platforms and the marketplace, thus impacting users’ privacy. In this paper, we propose the initial steps towards a change of paradigm, where the ratings are kept as private in each platform. Under this paradigm, each platform produces its rankings, then aggregated by the marketplace, in a federated fashion. To ensure that the marketplace’s rankings maintain their effectiveness, we exploit the concept of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">reputation of the individual platform</i>, so that the final marketplace ranking is weighted by the reputation of each platform providing its ranking. Experiments on three datasets, covering different use cases, show that our approach can produce effective rankings, improving robustness to attacks, while keeping user preference data private within each seller platform.