BROWSE
Detail
Conference Paper
Except-Condition Generative Adversarial Network for Generating Trajectory Data
Cited 3 time in 
Share 
Share
- Authors
- Yeji Song, Jihwan Shin, Jinhyun Ahn, Taewhi Lee, Dong-Hyuk Im
- Issue Date
- 2023-08
- Citation
- International Conference on Database and Expert Systems Applications (DEXA) 2023 (LNCS 14147), v.14147, pp.289-294
- Publisher
- Springer
- Language
- English
- Type
- Conference Paper
- Abstract
- Location data shared on social media is collected and processed as trajectory data, which exposes individuals to leakage of sensitive information, such as sensitive geographic areas. A typical countermeasure is a generative adversarial network (GAN) model that ensures data anonymity. However, generating data selectively by identifying only sensitive areas is difficult. In this study, we propose an except-condition GAN (exGAN) model that generates synthetic data while maintaining the original’s utility. This model ensures the anonymity of sensitive areas and maintains the distribution of data in relatively less sensitive areas. It uses a method that assigns the remaining labels except for specific selected labels as a condition. The selected labels represent points of high sensitivity, and the trajectory data generated by the model contains only points corresponding to the labels, except for the selected labels. Furthermore, in our comparative evaluation of the exGAN model, it outperformed the original GAN model.
- KSP Keywords
- Comparative Evaluation, High Sensitivity, Leakage of sensitive information, Location Data, Sensitive areas, Social media, Synthetic data, Trajectory Data, data location, generative adversarial network