Rethinking Fair Graph Neural Networks from Re-balancing

This is the official implementation of the following paper:

Rethinking Fair Graph Neural Networks from Re-balancing (SIGKDD'2024)

Table of Contents

• Rethinking Fair Graph Neural Networks from Re-balancing
  • Dependencies
  • Datasets
  • Running

Dependencies

• python>=3.7
• torch==2.0.1
• torch-geometric==2.3.1
• torch-scatter==2.1.1
• numpy==1.24.4
• scikit-learn==1.3.0

Datasets

We conduct experiments on three widely used real-world datasets, namely German Credit, Bail, and Credit Defaulter. The detailed information of the datasets is as follows. Please unzip datasets in ./dataset before running the model.

Dataset	German	Bail	Credit
# Nodes	1,000	18,876	30,000
# Edges	22,242	321,308	152,377
# Attributes	27	18	13
Sens.	Gender	Race	Age
Label	Credit status	Bail decision	Future default

Running

The run.sh includes details to reproduce experimental results in the paper:

bash run.sh

Acknowledgement

Our code is based on the FairVGNN (Improving fairness in graph neural networks via mitigating sensitive attribute leakage) and GraphENS (Graphens: Neighbor-aware ego network synthesis for class-imbalanced node classification).