Privacy-Preserving Personalized Recommender Systems

36 Pages Posted: 13 Sep 2022 Last revised: 8 Jun 2023

See all articles by Xingyu Fu

Xingyu Fu

Hong Kong University of Science & Technology (HKUST)

Ningyuan Chen

University of Toronto - Rotman School of Management

Pin Gao

School of Data Science, The Chinese University of Hong Kong, Shenzhen

Yang Li

University of Western Ontario - Richard Ivey School of Business

Date Written: August 27, 2022

Abstract

Problem Definition: Personalized product recommendations are essential for online platforms, but they raise privacy concerns due to the risk of inference attacks. To address this issue, we propose personalized recommendation policies that adhere to differential privacy constraints. Methodology and Results: We study a theoretical model where the recommendation policy selects products to recommend based on consumers’ preference rankings learned from personal data like cookies. By implementing differential privacy, we introduce randomness into the recommendation outcomes, preventing inference attacks. Our analysis shows that the optimal policy is a coarse-grained threshold policy, where products are randomly selected, with a subset having higher recommendation probabilities than the remaining options. The priority subset is determined by a threshold applied to the consumer’s preference ranking. We examine the choice of this threshold in the asymptotic regime with a large number of products, which is relevant to most online platforms. Additionally, we explore the economic implications of privacy protection. If product prices are exogenously determined, privacy protection reduces consumer surplus for sure due to the decreased match value of the recommended product. However, when retailers optimally set prices, we find that the impact of privacy protection on surplus is non-monotonic due to the trade-off between recommendation accuracy and price inflation. Managerial and Regulatory Implications: Our study offers valuable insights for practitioners seeking to design privacy-preserving personalized recommendation policies. We also provide regulators with a better understanding of the economic repercussions of privacy protection.

Keywords: differential privacy, personalization, product recommendations, socially-responsible operations

Suggested Citation

Fu, Xingyu and Chen, Ningyuan and Gao, Pin and Li, Yang, Privacy-Preserving Personalized Recommender Systems (August 27, 2022). Available at SSRN: https://ssrn.com/abstract=4202576 or http://dx.doi.org/10.2139/ssrn.4202576

Xingyu Fu (Contact Author)

Hong Kong University of Science & Technology (HKUST) ( email )

Clearwater Bay
Kowloon, 999999
Hong Kong

Ningyuan Chen

University of Toronto - Rotman School of Management ( email )

Pin Gao

School of Data Science, The Chinese University of Hong Kong, Shenzhen ( email )

Yang Li

University of Western Ontario - Richard Ivey School of Business ( email )

1151 Richmond Street North
London, Ontario N6A 3K7
Canada

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
618
Abstract Views
1,728
Rank
82,348
PlumX Metrics