Privacy-Preserving Collaborative Filtering System For Book-Crossing Dataset

Abstract

Web services that store and use their users’ sensitive data can cause privacy violation issues. Using personal preferences to generate predictions may increase individuals’ privacy risks in collaborative recommendation systems. Users who worry about privacy violations may be willing to provide false information and sometimes refuse to use these services. As a result, the recommender system’s prediction generation quality will decrease because it is an undeniable fact that the accuracy of prediction is directly related to the quality of the collected user data. It is crucial to discuss the privacy risks that may arise from the use of such systems and to protect user data privacy with accepted privacy protection mechanisms to alleviate user concerns. In this study, we evaluate the randomized perturbation-based privacy protection mechanism on a traditional memory-based collaborative filtering system that used the Book-Crossing dataset. We also compared recommendation accuracy over varying levels of privacy to find a balance between accuracy and privacy issues. Experimental results based on real-world user data show that a privacy-preserving scheme maintains the confidentiality of personal preferences without severely compromising prediction accuracy.