Privacy-preserving naïve bayesian classifier-based collaborative filtering

Collaborative filtering (CF) has become very popular on the Internet. Although CF systems are widely used, they have various challenges in recommendation process. The first one is collection of users? private data. For better referrals, such systems need quality data; however, due to privacy concerns, users hesitate to send their private data or they might decide to send false data. The second challenge is that CF systems provide referrals on existing databases compromised of ratings recorded from groups of people evaluating various items; sometimes, however, the systems? ratings might be split among different parties. The parties may wish to share their data; but they may not want to disclose their data. The third challenge is optimizing problem. Prediction process? run time increases with increasing number of users/items. It becomes difficult to offer referrals efficiently.In this thesis, approaches are proposed to overcome challenges for naïve Bayesian classifier (NBC)-based CF algorithm. A new scheme is proposed to produce NBC-based recommendations while preserving users? privacy by utilizing randomized response techniques (RRT). To offer CF services on distributed data between two parties without violating their privacy, solutions are provided. And finally, a method is proposed for optimizing privacy-preserving NBC-based CF scheme using k-modes clustering. To assess the proposed schemes, experiments are conducted using real data sets. The solutions are analyzed in terms of accuracy, privacy, and additional costs. After drawing conclusions, future works are presented.