CRoC: Context Refactoring Contrast for Graph Anomaly Detection with Limited Supervision

This is the code for reproducing the results reported in the ECAI2025 paper.

More details of the code will be updated soon.

Requirements:

• PyTorch
• DGL
• Numpy
• Scipy
• scikit-learn

Quick Start

For quick reproduction of the CRoC(GIN) results in Yelp, you can type the following command in the terminal:

python main.py

Reproduce Results

To reproduce other reported results, you can specify the dataset, model and hyper-parameters, e.g., reproduce the results on Amazon:

python main.py --model CRoCSAGE --dataset amazon --alpha 0.5 --gamma 0.2 --eta 0.5 --n_epoch 200

Note that Yelp and Amazon can be downloaded through DGL, while T-Soc, T-Fin and DGraph-Fin should be manually downloaded and placed under the ./dataset folder.

You can download these three datasets via:

• T-Soc and T-Fin: https://github.com/squareRoot3/Rethinking-Anomaly-Detection
• DGraph: https://dgraph.xinye.com/dataset

Hyper-parameters of experiments in each dataset is provided in Table 10 of the arxiv paper.

Citation

@article{xie2025croc,
  title={CRoC: Context Refactoring Contrast for Graph Anomaly Detection with Limited Supervision},
  author={Xie, Siyue and Tam, Da Sun Handason and Lau, Wing Cheong},
  journal={arXiv preprint arXiv:2508.12278},
  year={2025}
}