DIAM

This repository is an official PyTorch(Geometric) implementation of DIAM in "Effective Illicit Account Detection on Large Directed MultiGraph Transaction Networks of Cryptocurrencies".

Requirements

• Python==3.9
• torch==1.12.0+cu116
• torch-geometric==2.2.0
• scikit-learn==1.1.1
• scipy==1.8.1
• numpy==1.22.4
• dgl-cu116==0.8.2.post1

Data

Dataset	#Nodes	#Edges	#Edge attribute	#Illicit	#Normal	Illicit:Normal
Ethereum-S	1,329,729	6,794,521	2	1,660	1,700	1:1.02
Ethereum-P	2,973,489	13,551,303	2	1,165	3,418	1:2.93
Bitcoin-M	2,505,841	14,181,316	5	46,930	213,026	1:4.54
Bitcoin-L	20,085,231	203,419,765	8	362,391	1,271,556	1: 3.51

We provide two ways to obtain the datasets:

Option 1: Original distribution (SharePoint)

Download the four datasets from the authors' original link, then extract them into the ./data folder:

• datasets (SharePoint)

After extraction, follow the usage steps below (e.g., preprocess.py, then main.py).

Option 2: 🤗 Hugging Face mirror (.npz graph dict)

We also mirror the datasets on Hugging Face in a stable .npz format, where each file stores a single large graph as a graph dictionary:

• Hugging Face dataset repo:
  https://huggingface.co/datasets/Tommy-DING/crypto-illicit-account-detection-multigraphs

Files provided:

• EthereumS_graph_dict.npz
• EthereumP_graph_dict.npz
• BitcoinM_graph_dict.npz
• BitcoinL_graph_dict.npz

Each *_graph_dict.npz includes:

• edge_index: shape [2, E] (COO edge list)
• edge_attr: shape [E, D] (edge features)
• X: shape [N, F] (node features)
• y: shape [N] with labels in {-1, 0, 1} (-1: unknown, 0: benign, 1: illicit)

Note: This repository (DIAM) uses PyTorch Geometric. If you prefer the original .pt files and the official end-to-end pipeline, please use this GitHub repo directly.
The Hugging Face version is intended for easier downloading and broader interoperability.

Usage

1. Download and extract data files to /data folder.

2. Generate edge sequences.

   python3 preprocess.py

3. Run DIAM.

   python3 main.py

Citation

@inproceedings{ding2024effective,
  title={Effective illicit account detection on large cryptocurrency multigraphs},
  author={Ding, Zhihao and Shi, Jieming and Li, Qing and Cao, Jiannong},
  booktitle={Proceedings of the 33rd ACM International Conference on Information and Knowledge Management},
  pages={457--466},
  year={2024}
}