Behavior-aware Account De-anonymization on Ethereum Interaction Graph

This is a Python implementation of Ethident, as described in the following:

Behavior-aware Account De-anonymization on Ethereum Interaction Graph

Requirements

For hardware configuration, the experiments are conducted at Ubuntu 18.04.5 LTS with the Intel(R) Xeon(R) Gold 5218R CPU @ 2.10GHz, and NVIDIA Tesla V100S GPU (with 40GB memory each). For software configuration, all model are implemented in

• Python 3.7
• Pytorch-Geometric 2.0.3
• Pytorch 1.8.0
• Scikit-learn 0.24.1
• CUDA 10.2

Data

Download data in PYG format from this page and place it under the 'data/' path.

Note that we store the raw block data (downloaded from the xblock platform) in the neo4j database, which is huge, so we are not ready to publish it. You can download the raw block data from the xblock platform.

Usage

Execute the following bash commands in the same directory where the code resides:

$ python main_ggc.py -l i --hop 2 -ess Volume -layer 2 --pooling max --hidden_dim 128 --batch_size 32 --lr 0.001 --dropout 0.2 -undir 1 --aug edgeRemove+identity --aug_prob1 0.1

More parameter settings can be found in 'utils/parameters.py'.

Citation

If you find this work useful, please cite the following:

@article{zhou2022behavior,
  title={Behavior-aware account de-anonymization on ethereum interaction graph},
  author={Zhou, Jiajun and Hu, Chenkai and Chi, Jianlei and Wu, Jiajing and Shen, Meng and Xuan, Qi},
  journal={IEEE Transactions on Information Forensics and Security},
  volume={17},
  pages={3433--3448},
  year={2022},
  publisher={IEEE}
}