Privacy-Preserving Ridesharing Recommendation in Geosocial Networks

Abstract

Geosocial networks have received a lot of attentions recently and enabled many promising applications, especially the on-demand transportation services that are increasingly embraced by millions of mobile users. Despite the well understood benefits, such services also raise unique security and privacy issues that are currently not very well investigated. In this paper, we focus on the trending ridesharing recommendation service in geosocial networks, and propose a new privacy-preserving framework with salient features to both users and recommendation service providers. In particular, the proposed framework is able to recommend whether and where the users should wait to rideshare in given geosocial networks, while preserving user privacy. Meanwhile, it also protects the proprietary data of recommendation service providers from any unauthorised access, such as data breach incidents. These privacy-preserving features make the proposed framework especially suitable when the recommendation service backend is to be outsourced at public cloud for improved service scalability. On the technical front, we first use kernel density estimation to model destination distributions of taxi trips for each cluster of the underlying road network, denoted as cluster arrival patterns. Then we utilize searchable encryption to carefully protect all the proprietary data so as to allow authorised users to retrieve encrypted patterns with secure requests. Given retrieved patterns, the user can safely compute the potential of ridesharing by investigating the probabilities of possible destinations from ridesharing requirements. Experimental results show both the effectiveness of the proposed recommendation algorithm comparing to the naive “wait-at-where-you-are” strategy, and the efficiency of the utilized privacy-preserving techniques.