Scholarlink

The continuous deepening of mobile Internet and positioning technology has promoted the wide application of trajectory data in fields such as smart cities and transportation planning. However, the spatio-temporal behavior patterns it contains pose significant risks of privacy leakage. To address this, this paper proposes Tran-TrajGAN, a Transformer-based trajectory generative adversarial model. The model utilizes a Transformer encoder to model long-range spatiotemporal dependencies, incorporates local convolutional modules to extract fine-grained features, and designs a dynamic gating mechanism to adaptively adjust noise intensity, thereby preserving privacy while maintaining data utility. The study evaluates the privacy and utility of synthetic trajectories using Trajectory-User Linking (TUL) and Trajectory Sharing Percentage (TSP) metrics on Geolife and Foursquare NYC datasets. Experimental results demonstrate that Tran-TrajGAN outperforms other models in effectively balancing privacy protection and data utility.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Transformer-Based Dynamic Gated Trajectory Privacy-Preserving Generative Adversarial Model

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yangdingkang Huang,
  • Zhiguang Chu,
  • Tian Sang,
  • Yang Gao,
  • Xing Zhang

摘要

The continuous deepening of mobile Internet and positioning technology has promoted the wide application of trajectory data in fields such as smart cities and transportation planning. However, the spatio-temporal behavior patterns it contains pose significant risks of privacy leakage. To address this, this paper proposes Tran-TrajGAN, a Transformer-based trajectory generative adversarial model. The model utilizes a Transformer encoder to model long-range spatiotemporal dependencies, incorporates local convolutional modules to extract fine-grained features, and designs a dynamic gating mechanism to adaptively adjust noise intensity, thereby preserving privacy while maintaining data utility. The study evaluates the privacy and utility of synthetic trajectories using Trajectory-User Linking (TUL) and Trajectory Sharing Percentage (TSP) metrics on Geolife and Foursquare NYC datasets. Experimental results demonstrate that Tran-TrajGAN outperforms other models in effectively balancing privacy protection and data utility.