Paper 2021/006

Privacy-Preserving Privacy Profile Proposal Protocol

Wyatt Howe and Andrei Lapets

Abstract

Many web-based and mobile applications and services allow users to indicate their preferences regarding whether and how their personal information can be used or reused by the application itself, by the service provider, and/or by third parties. The number of possible configurations that constitute a user's preference profile can be overwhelming to a typical user. This report describes a practical, privacy-preserving technique for reducing the burden users face when specifying their preferences by offering users data-driven recommendations for fully-specified preference profiles based on their inputs for just a few settings. The feasibility of the approach is demonstrated by a browser-based prototype application that relies on secure multi-party computation and uses the web-compatible JIFF library as the backbone for managing communications between the client application and the recommendation service. The principal algorithms used for generating proposed preference profiles are $k$-means clustering (for privacy-preserving analysis of preference profile data across multiple users) and $k$-nearest neighbors (for selecting a proposed preference profile to recommend to the user).