NodeImport: Imbalanced Node Classification with Node Importance Assessment

This is the code for the KDD 2025 paper of NodeImport: NodeImport: Imbalanced Node Classification with Node Importance Assessment.

Overview

Key components of the repository:

• Main Training Procedure: Implemented in main.py.
• Importance Calculation: Implemented in calculate_importance in imp_calc.py.

Requirements

To install the necessary packages, run:

pip install -r requirements.txt

Datasets

• We use the built-in datasets from the PyTorch Geometric package.
• Datasets will be downloaded into the default data folder upon first execution.

Running

Hyper-parameter configurations for all Dataset+GNN settings are available in the run.sh file. To run the model as described in the paper, specify the required hyper-parameters and run the following command:

./run.sh [dataset] [GNN backbone] [imbalance ratio] [running times] [GPU device]

For instance, to run the model on the Cora dataset with a GCN backbone, an imbalance ratio of 50, for 5 runs on GPU device 4, execute:

./run.sh Cora GCN 50 5 4

Acknowledgments

The implementation of GNN backbones in the nets folder is adapted from TAM. We thank the authors of TAM for their valuable contributions.

Citation

If you find our work useful, please cite:

Nan Chen, Zemin Liu, Bryan Hooi, Bingsheng He, Jun Hu, and Jia Chen. 2025. NodeImport: Imbalanced Node Classification with Node Importance Assessment. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1 (KDD ’25), August 3–7, 2025, Toronto, ON, Canada. ACM, New York, NY, USA, 12 pages. https://doi.org/10.1145/3690624.3709215

Feel free to contact [email protected] if you have any questions.