Privacy-preserving top-N recommendation on distributed data

Abstract

Traditional collaborative filtering (CF) systems perform filtering tasks on existing databases; however, data collected for recommendation purposes may split between different online vendors. To generate better predictions, offer richer recommendation services, enhance mutual advantages, and overcome problems caused by inadequate data and/or sparseness, e-companies want to integrate their data. Due to privacy, legal, and financial reasons, however, they do not want to disclose their data to each other. Providing privacy measures is vital to accomplish distributed databased top-N recommendation (TN), while preserving data holders' privacy. In this article, the authors present schemes for binary ratings-based TN on distributed data (horizontally or vertically), and provide accurate referrals without greatly exposing data owners' privacy. Our schemes make it possible for online vendors, even competing companies, to collaborate and conduct TN with privacy, using the joint data while introducing reasonable overhead costs.