Deep Learning-based Privacy-preserving Publishing Method for Location Big Data in Vehicular Networks

In contemporary times, there is an increasing integration of Location-Based Service (LBS) enabled smart devices into the fabric of individuals' daily lives. The prevalent era of large-scale models predicting users' historical location points poses a significant threat to user privacy. Simultaneously, conventional data release models exhibit suboptimal performance. This paper proposes a novel approach incorporating a deep learning prediction model and a location data release method called Hilbert-ConvLSTM, aimed at enhancing data availability while ensuring the privacy of user information. Firstly, leveraging the properties of the Hilbert curve, the predicted location point data is partitioned into multiple spatio-temporal structures. A sampling mechanism and exponential mechanism are employed for the selection of representative points within each location cluster. Subsequently, utilizing the "4V" characteristics of location point data, deep learning models are employed to extract spatio-temporal features, facilitating the prediction of location point data. Finally, in conjunction with the architecture derived from Hilbert curve partitioning, differential privacy budget allocation and Laplace noise addition are applied to achieve privacy protection in the statistical partitioned release of large-scale location data. Experimental analyses using real-world data validate the proposed method's advantages in terms of data release usability and efficiency.