Effects of Inconsistently Masked Data Using RPT on CF with Privacy

Abstract

Randomized perturbation techniques (RPT) are applied to perturb the customers' private data to protect privacy while providing accurate referrals. In the RPT-based collaborative filtering (CF) with privacy schemes, proposed so far, users disguise their ratings in the same way to achieve consistently perturbed data. However, since users might have different levels of concerns about their privacy, the customers might decide to perturb their private data differently, which causes inconsistently masked data. How, then, can e-companies present referrals using such data and how can inconsistent data disguising affect accuracy and privacy?