Title: Privacy-preserving item-based recommendations over partitioned data with overlaps

Authors: Ibrahim Yakut; Jaideep Vaidya

Addresses: Department of Computer Engineering, Anadolu University, Iki Eylul Campus, Eskisehir, 26555, Turkey ' MSIS Department, Rutgers University, 1 Washington Park, Newark, NJ 07102, USA

Abstract: User ratings are vital elements to drive recommender systems and, in the case of an insufficient amount of ratings, companies may prefer to operate recommender services over partitioned data. To make this feasible, there are privacy-preserving schemes. However, such solutions currently have not comprehensively investigated probable rating overlaps among partitioned data. Such overlaps make collaboration over partitioned data more challenging, especially if overlapped values are divergent. In this study, we examine this privacy-preserving recommender problem and propose novel schemes in this sense. By means of our schemes, two parties can perform item-based collaborative filtering over partitioned data with divergent overlaps. We also show that the proposed solutions promote prediction quality with tolerable overheads.

Keywords: collaborative filtering; privacy; arbitrary partitioning; rating overlaps.

DOI: 10.1504/IJBIS.2017.084449

International Journal of Business Information Systems, 2017 Vol.25 No.3, pp.336 - 351

Available online: 12 May 2017 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article