A Privacy Review of Vertically Partitioned Data-based PPCF Schemes

Abstract

E-commerce companies utilize collaborative filtering approaches to provide recommendation in order to attract customers. Consumer participation through supplying feedback is an important component for a recommendation system to produce accurate predictions. New companies in the marketplace might lack enough data for collaborative filtering purposes. Thus, companies can come together to share their horizontally or vertically partitioned data for better services. Although partitioned data-based recommendation schemes provide accurate predictions, privacy issues might pose different risks to the companies participating into such collaboration. Privacy-preserving collaborative filtering schemes aim to provide accurate predictions without neglecting the privacy of such data holders. However, the collaborating parties’ privacy might not be protected as much as believed provided by these schemes. In this study, the privacy, offered by vertically partitioned binary ratings-based privacy-preserving collaborative filtering schemes, is examined by three different attacks and experimentally tested. Empirical outcomes show that the collaborating parties are still able to derive their confidential data.