Home

Awesome

Introduction

This is the code for the ICLR 2024 paper of PMP: Partitioning Message Passing for Graph Fraud Detection.

Label imbalance and homophily-heterophily mixture are the fundamental problems encountered when applying Graph Neural Networks (GNNs) to Graph Fraud Detection (GFD) tasks. In this work, we introduce Partitioning Message Passing (PMP), an intuitive yet effective message passing paradigm expressly crafted for GFD. In the neighbor aggregation stage of PMP, neighbors with different classes are aggregated with distinct node-specific aggregation functions. By this means, the center node can adaptively adjust the information aggregated from its heterophilic and homophilic neighbors, thus avoiding the model gradient being dominated by benign nodes which occupy the majority of the population. A brief overview of our framework is illustrated in the following picture.

<p align="center"><img src="pic/framework.png" alt="logo" width="800px" /></p>

Directory Structure

The repository is organized into several directories, each serving a specific purpose:

Installation

Usage:

Pretrained Model

Run our pretrained model to reproduce the results provided in the paper

python test.py --dataset yelp/amazon/tfinance

Running Example

<p align="center"><img src="pic/example.png" alt="logo" width="800px" /></p>

Citation

If you find our work useful, please cite:

@inproceedings{
zhuo2024partitioning,
title={Partitioning Message Passing for Graph Fraud Detection},
author={Wei Zhuo and Zemin Liu and Bryan Hooi and Bingsheng He and Guang Tan and Rizal Fathony and Jia Chen},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024},
url={https://openreview.net/forum?id=tEgrUrUuwA}
}

Feel free to contact jhuow@proton.me if you have any questions.