Privacy-preserving collaborative filtering

Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering and Computer Science


Wenliang Du


Collaborative filtering, Privacy, Randomization, Partitioned data

Subject Categories

Computer Sciences


With the evolution of the Internet, collaborative filtering (CF) techniques are becoming increasingly popular. Such techniques are widely used by many E-commerce companies to suggest products to customers, based on like-minded customers' preferences; but they fail to protect users' privacy. Data from customers is gathered for recommendation purposes; however, collecting high quality data is not an easy task, due to privacy concerns. Since many users are worried about their privacy, they sometimes refuse to contribute their data; or occasionally they might decide to give false information. Producing accurate referrals based on insufficient data and/or false data is impossible. CF systems provide referrals on existing databases comprised of ratings recorded from groups of people evaluating various items; sometimes, however, the systems' ratings are split among different parties. To provide better filtering services, the parties may wish to share their data; but they may not want to disclose information of a private nature. We propose to use randomization techniques to protect users' privacy while still producing accurate referrals. We investigate providing predictions on memory- and model-based CF algorithms with privacy assured by using randomized perturbation techniques (RPT) in Chapter 3 and Chapter 4, respectively; while we show how to produce recommendations without violating users' privacy using randomized response techniques (RRT) in Chapter 5. In Chapter 6, we explore inconsistently disguised data-based CF with privacy provided by the RPT. When data owners wish to provide recommendations using a joint database without disclosing sensitive data to each other, they may achieve accurate privacy-preserving referrals on integrated data with schemes we have proposed. We investigate performing filtering tasks; including prediction generations and top- N recommendations, on vertically or horizontally partitioned data, with privacy preservation, in Chapter 7 and Chapter 8. We conduct various experiments using well-known existing data sets to evaluate the overall performance of our approaches. Furthermore, we analyze the effects of different parameters on accuracy and privacy. Finally, in Chapter 9, we present our results and explain future directions.


Surface provides description only. Full text is available to ProQuest subscribers. Ask your Librarian for assistance.