Privacy-preserving dimensionality reduction-based collaborative filtering

Collaborative filtering (CF) systems are widely used by many e-commerce sites. However, they fail to provide privacy measures. A significant amount of internet users do not feel comfortable to give their data for CF purposes due to privacy concerns. That is why it becomes a challenge to collect truthful and dependable data to perform CF services. Researches show that privacy concerns differ from user to user. Therefore, users might decide to hide their private data differently. Providing CF services on variably masked data is challenging. Two parties may need to combine their data for CF purposes for better recommendations. However, they do not want to integrate them due to privacy, legal, and financial reasons. If privacy measures are provided, they can combine their data. The question is then how they can offer CF services on integrated data without violating their privacy.In this thesis, first, solutions are proposed to offer CF services based on Eigentaste algorithm without violating individual users? privacy. Second, it is shown how to provide recommendations using singular value decomposition (SVD)-based algorithms from inconsistently perturbed data. Finally, it is investigated how to achieve SVD-based CF on distributed data between two parties while preserving their privacy. To evaluate the overall performance of the proposed solutions, experiments are conducted using real data sets collected for CF purposes. The proposed schemes are analyzed in terms of accuracy, privacy, and additional costs. After explaining the solutions, conclusions are drawn and future directions are presented.