Privacy Preservation for Trajectory Publication Based on Differential Privacy
Lin Yao,Zhenyu Chen,Haibo Hu,Guowei Wu,Bin Wu
DOI: https://doi.org/10.1145/3474839
IF: 5
2022-06-30
ACM Transactions on Intelligent Systems and Technology
Abstract:With the proliferation of location-aware devices, trajectory data have been used widely in real-life applications. However, trajectory data are often associated with sensitive labels, such as users’ purchase transactions and planned activities. As such, inappropriate sharing or publishing of these data could threaten users’ privacy, especially when an adversary has sufficient background knowledge about a trajectory through other data sources, such as social media (check-in tags). Though differential privacy has been used to address the privacy of trajectory data, no existing method can protect the privacy of both trajectory data and sensitive labels. In this article, we propose a comprehensive trajectory publishing algorithm with three effective procedures. First, we apply density-based clustering to determine hotspots and outliers and then blur their locations by generalization. Second, we propose a graph-based model to efficiently capture the relationship among sensitive labels and trajectory points in all records and leverage Laplace noise to achieve differential privacy. Finally, we generate and publish trajectories by traversing and updating this graph until we travel all vertexes. Our experiments on synthetic and real-life datasets demonstrate that our algorithm effectively protects the privacy of both sensitive labels and location data in trajectory publication. Compared with existing works on trajectory publishing, our algorithm can also achieve higher data utility.
computer science, information systems, artificial intelligence