Randomized Algorithms for Precise Measurement of Differentially-private, Personalized

This paper was accepted at The fifth AAAI Workshop on Privateness-Preserving Synthetic Intelligence.

Customized suggestions type an essential a part of right this moment’s web ecosystem, serving to artists and creators to achieve customers, and serving to customers to find new and fascinating content material. Nevertheless, many customers right this moment are skeptical of platforms that personalize suggestions, partially because of traditionally careless remedy of non-public information and information privateness. Now, companies that depend on customized suggestions are coming into a brand new paradigm, the place lots of their programs have to be overhauled to be privacy-first. On this article, we suggest an algorithm for customized suggestions that facilitates each exact and differentially-private measurement. We think about promoting for example utility, and conduct offline experiments to quantify how the proposed privacy-preserving algorithm impacts key metrics associated to consumer expertise, advertiser worth, and platform income in comparison with the extremes of each (non-public) non-personalized and non-private, customized implementations.