BTG

PyTorch implementation for the paper below:
BTG: A Bridge to Graph machine learning in telecommunications fraud detection

Running BTG

To run the code, you need to have at least Python 3.7 or later versions. And follow the steps below :
1.Go to this site to download the 4 datasets, namely train_app.csv,train_sms.csv,train_user.csv,train_voc.csv;
2.Put the 4 downloaded datasets in the path: /BTG/data/train;
3.Run python data_process.py to extract features and generate dataset in DGL;
4.Run python train.py to run BTG with default settings on the dataset.

Repo Structure

The repository is organized as follows:

• data_process.py: convert raw node features and adjacency matrix to DGL dataset;
• main.py: training and testing BTG;
• model.py: BTG model implementations;
• layers.py: model layers;
• utils.py: utility functions;
• data: raw dataset in /data/train, and extracted dataset in /data/user_data.

Citation

@article{hu2022btg,
  title={BTG: A Bridge to Graph machine learning in telecommunications fraud detection},
  author={Hu, Xinxin and Chen, Hongchang and Liu, Shuxin and Jiang, Haocong and Chu, Guanghan and Li, Ran},
  journal={Future Generation Computer Systems},
  volume={137},
  pages={274--287},
  year={2022},
  publisher={Elsevier}
}