DHMP

This is the PyTorch implementation for the Dual-channel Heterophilic Message Passing for Graph Fraud Detection, which was accepted by IJCNN 2025.

Data

Datasets and the usage can be found in SplitGNN: Spectral Graph Neural Network for Fraud Detection against Heterophily, we also thank for their great work.

Usage

• src/: includes all code scripts.
• data/: includes original datasets:
  • YelpChi.zip: The original dataset of YelpChi, which contains hotel and restaurant reviews filtered (spam) and recommended (legitimate) by Yelp.
  • Amazon.zip: The original dataset of Amazon, which contains product reviews under the Musical Instruments category.
  • FDCompCN.zip: The processed dataset of FDCompCN, which contains financial statement fraud of companies in China from CSMAR database.
• config/: includes the setting of parameters for two datasets.
  • yelp.yaml: The general parameters of YelpChi.
  • amazon.yaml: The general parameters of Amazon.
  • comp.yaml: The general parameters of FDCompCN.
• result/: includes the results of models.

Model Training

python train.py --dataset yelp