Tradeoff between Privacy and Utility for Location-based Recommendation Services

Abstract

Location-based recommendation services (LBRS) are widely used by people to find new places of interest. However, the prevalence of LBRS poses a severe threat to users’ privacy, because LBRS queries contain sensitive information such as users’ preferences and location. Many Location Privacy Protection Mechanisms (LPPMs) have been proposed to mitigate the threat by obfuscating actual location in the query with a noise-based privacy-preserving technique. However, disguised location information can potentially harm the user experience with LBRS. In this work, we evaluate the impact of the noise-based privacy-preserving technique on user-perceived utility measured as nDCG (normalized Discounted Cumulative Gain) ranks. We empirically evaluate user-perceived utility under different noise levels to explore the trade-off between privacy and utility. A variety of factors, including service density and mechanism employed by service providers, are found to impact the utility loss, including but not limited to different service providers, different service types, and different sorts of areas (e.g., urban v.s. rural).