Protecting personal trajectories of social media users through differential privacy

Abstract

Road traffic congestion is an important issue in modern cities, however most existing traffic jam identification solutions are based on expensive facilities such as sensors or transport probe infrastructure with high deployment and management costs. As a result of the cost, such solutions are not ubiquitously deployed. The extensive use of smart mobile devices furnished with location-based competencies and the global popularity of microblogging applications like Twitter offers an opportunity to tackle these problems. Twitter users can serve as human traffic sensors providing real-time reflections of current traffic situations. However, these data can contain extensive personal privacy information that demands privacy preserving mechanisms for the user location and their current trajectory. Differential privacy can ensure that degrees of privacy in these trajectories can be preserved whilst allowing data analysis and mining of the Twitter content. This paper proposes an innovative private trajectories release model and associated algorithms with differential privacy guarantees that considers both data privacy and data utility. This includes development of a private reference system for calibrating separate (raw) users trajectories across obfuscated anchor points; construction of privacy supporting noise-enhanced prefix trees to release synthesis data privately, and comprehensive evaluation of both the accuracy and utility of the solutions in terms of a set of evaluation metrics based on real-life tweets-based user trajectories across the city of Melbourne.

Introduction

With the extensive use of smart mobile devices furnished with location-based competencies and the global popularity of micro-blogging applications like Twitter, new possibilities now exist for collection of near real-time transport data. Tweets, especially generated by car drivers, passengers and other pedestrians near roads, can be regarded as indicators of real-time traffic conditions. This however raises privacy issues. It is vital to perform behavioural identification and pattern mining over trajectories to provide traffic monitoring information with individual privacy protected, particularly when it might be used for malicious purposes against Tweeters. Differential privacy (Dwork et al. 2006) has been broadly adopted to protect the privacy of sensitive data including location and trajectory data. It can rigorously protect the privacy of individual location and trajectory information whilst still providing synthesised data for analysis. Approaches that can limit the dangers of leaking location privacy would encourage more users to share their location information. However, when spatial temporal data in social media are released and used to aid in monitoring traffic congestion, there are several challenges that emerge due to the heterogeneity and complex nature of trajectories. Firstly, de Montjoye et al. (2013) showed that over 90% of trajectories can be identified using no more than 4 given locations. Furthermore, coarse and fine sample approaches lead to significantly different results. Consequently, it is crucial to propose privacy preserving trajectory publication solutions for social media under differential privacy mechanisms whilst supporting the use of such data more generally, e.g. for congestion identification.

The goal of this paper is to support a privacy supporting trajectory calibration and publication system leveraging differential privacy as the basis for social media location data obfuscation and protection. Existing research works on trajectories with privacy guarantees have several restrictions and drawbacks. They are imperfect for sampling raw trajectories using fixed time or space intervals. Most real life evaluations privacy approaches only address short trajectories across a small quantity of locations (e.g. metro stations) in restricted space areas and the reference systems adopted for sampling raw trajectories are often inefficient and without privacy guarantees. Chen et al. (2012a) and Chen et al. (2013) provided a non-interactive synthesis of trajectories (sequence) with differential privacy guarantees but restricted their solution to short sequences across metro stations. When dealing with larger scale space areas and more realistic scenarios, e.g. trajectories analysis over traffic in cities, this approach is ineffective. Moreover, it is insufficient to build static noisy prefix trees (modelling trajectories), since there is a need for differential privacy guarantees as the number of nodes in the tree increases. Chen et al. (2012a) used an n-gram approach to improve this, but this led to trajectory and sequence information loss. Su et al. (2013) proposed a trajectory calibrating solution to publish trajectories, but without privacy preserving mechanisms. The model proposed here adopts a new privacy and accuracy-preserving reference system to extract privacy-demanding feature-based anchors which can subsequently be used to calibrate sequences from raw trajectories. This provides a private trajectory data sensitisation approach that scales to large spatial domains reflecting realistic trajectories as it occurs over large-scale real life social media data. The main contributions of this work are: (1) development of a privacy preserving reference system that can be used for protecting arbitrary trajectories, (2) improved utility of synthesised data under differential privacy guarantees, and (3) support for a privacy-preserving trajectory data publication solution that can scale to large space-time domains that reflect realistic trajectories that arise in large-scale social networks.

The rest of this paper is structured as follows. Section 2 describes the related works with focus on those that explore the advantages of differential privacy compared to other approaches used for protecting trajectories. Section 3 introduces the preliminary concepts used in the work as well as providing an overview of the solution. Section 4 presents the Private Reference System adopted in the work. Section 5 presents the Private Trajectories Release that has been adopted. The threat model and assumptions are defined in Section 6 and privacy analysis is performed as well. Section 7 presents the experimental results of the privacy-preserved trajectories using Twitter data focused especially on traffic events identified through social media. Finally Section 8 draws conclusions on the work as a whole and outlines areas of future work.