Mitigating Homophily Disparity in Graph Anomaly Detection: A Scalable and Adaptive Approach

Official Implementation of Mitigating Homophily Disparity in Graph Anomaly Detection: A Scalable and Adaptive Approach.

Getting Started

Setup Environment

To run the code, please install the following libraries: dgl==2.4.0+cu124, torch==2.4.0, numpy==2.1.3, scipy==1.14.1

Preparing Datasets

We use ten datasets provided by GADBench. After downloading, unzip all the files into the datasets folder.

Due to the Copyright of DGraph-Fin and Elliptic, you need to download these datasets by yourself. The script to preprocess DGraph-Fin and Elliptic can be found in GADBench/preprocess.ipynb. You can also preprocess your own dataset according to the notebook.

Model Configuration

Model configurations/hyperparameters are provided in the semi_train.conf.yaml.

Training and Evaluation

python main.py --dataset reddit

Acknowledgements

The code is implemented based on GADBench, UniGAD, and PolyGCL.