LSTM-TrajGAN: A Deep Learning Approach to Trajectory Privacy Protection
If you find our code useful for your research, please cite our paper:
Rao, J., Gao, S.*, Kang, Y. and Huang, Q. (2020). LSTM-TrajGAN: A Deep Learning Approach to Trajectory Privacy Protection. In the Proceedings of the 11th International Conference on Geographic Information Science (GIScience 2021), pp. 1-16.
LSTM-TrajGAN uses the following packages with Python 3.6.3
- numpy==1.18.4
- pandas==1.1.5
- tensorflow-gpu==1.13.1
- Keras==2.2.4
- geohash2==1.1
- scikit-learn==0.23.2
Train the LSTM-TrajGAN model using the preprocessed data.
python train.py 2000 256 100
Where 2000 is the total training epochs, 256 is the batch size, 100 is the parameter saving interval (i.e., save params every 100 epochs).
Generate synthetic trajectory data based on the real test trajectory data and save them to results/syn_traj_test.csv.
python predict.py 1900
Where 1900 means we load the params file saved at the 1900th epoch to generate synthetic trajectory data.
Evaluate the synthetic trajectory data on the Trajectory-User Linking task using MARC.
python TUL_test.py data/train_latlon.csv results/syn_traj_test.csv 100
Where data/train_latlon.csv is the training data, results/syn_traj_test.csv is the synthetic test data, 100 is the embedder size.
The data we used in our paper originally come from the Foursquare NYC check-in dataset.
We mainly referred to these two works:
May Petry, L., Leite Da Silva, C., Esuli, A., Renso, C., and Bogorny, V. (2020). MARC: a robust method for multiple-aspect trajectory classification via space, time, and semantic embeddings. International Journal of Geographical Information Science, 34(7), 1428-1450. Github
Keras-GAN: Collection of Keras implementations of Generative Adversarial Networks (GANs). Github