Node Duplication Improves Cold-start Link Prediction

Introduction

This repository contains the source code for the paper: Node Duplication Improves Cold-start Link Prediction

Requirements

The code was developed and tested with Python 3.10.9. The enviromental requirements are listed as in requirements.txt. Please run the following code to install all the requirements:

pip install -r requirements.txt

Prepare the datasets

IGB-100K dataset needs to be pre-downloaded and pre-processed. Please install IGB package and download IGB-100k datset as follows:

git clone https://github.com/IllinoisGraphBenchmark/IGB-Datasets.git
cd IGB-Datasets/
pip install .
cd IGB-Datasets
python
from igb import dataloader
from igb import download
download.download_dataset(path='../IGB-Datasets', dataset_type='homogeneous', dataset_size='tiny')

Then pre-process IGB-100K datasets with the following command:

cd data
python igb_process.py

Other datasets can be directly downloaded when running the experimental code.

Usage

Transductive Setting

Please run the following command to reproduce the results in Table 1:

bash scripts/transductive.sh

Production Setting

Please run the following command to reproduce the results in Table 2 and Table 7 in the paper:

bash scripts/inductive.sh

Citation

If you use this code in your research, please cite the following paper:

@article{guo2024node,
  title={Node Duplication Improves Cold-start Link Prediction},
  author={Guo, Zhichun and Zhao, Tong and Liu, Yozen and Dong, Kaiwen and Shiao, William and Shah, Neil and Chawla, Nitesh V},
  journal={arXiv preprint arXiv:2402.09711},
  year={2024}
}