Title: GAN-based Differential Privacy Trajectory Data Publishing with Sensitive Label
Abstract: With the development of location awareness technology, trajectory data is easy to be collected and applied to urban transportation, mobile medical care, etc. However, trajectory data is usually collected with some sensitive attributes such as disease and age, which may contain a large amount of users' private information. If these attributes are directly published without any privacy preservation, there will be a risk of privacy leakage. Differential privacy as a robust privacy protection model independent of any background knowledge attack has been widely used to protect trajectory data privacy and reduce privacy leakage. However, there are still some problems to be solved. On the one hand, most existing schemes based on differential privacy only can anonymize the trajectory data without considering sensitive labels (sensitive attributes). On the other hand, unified noise is added to the trajectory data even though some trajectory points may not cause any leakage, which reduces the overall utility of published data conversely. In this paper, we aim to achieve the trajectory privacy based on differential privacy by adopting Generative Adversarial Network (GAN). In our scheme, sensitive locations including key stop points and abnormal points which may cause the privacy disclosure will be determined. We design a graph model to effectively capture the relationship between the sensitive attributes and the corresponding sensitive points. Laplace noise is added to the graph, while the privacy budget is trained by GAN to balance data privacy and data utility. Experiments on real data sets can verify that our proposed scheme can achieve more privacy while possessing better utility.
Publication Year: 2022
Publication Date: 2022-08-01
Language: en
Type: article
Indexed In: ['crossref']
Access and Citation
Cited By Count: 1
AI Researcher Chatbot
Get quick answers to your questions about the article from our AI researcher chatbot