Note

This repository includes the implementation for our NeurIPS 2023 paper: Cross-links Matter for Link Prediction: Rethinking the Debiased GNN from a Data Perspective.

Environments

Python 3.7.6

Packages:

dgl_cu102==0.9.1.post1
numpy==1.19.2
python_louvain==0.15
networkx==2.5
tqdm==4.62.3
torch==1.12.1+cu102
community==1.0.0b1
dgl==1.1.0
PyYAML==6.0

community is an essential package to deploy the Louvain algorithm used in our work.

Run the following code to install all required packages.

> pip install -r requirements.txt

Note

We notice that now dgl_cu102==0.9.1.post1 can not be installed by conda/pip directly. One can refer to the previous packages and download the corresponding package before installation.

Datasets & Processed files

• Due to size limitation, the processed files and datasets are stored in google drive. The datasets include Epinions, DBLP and LastFM.
• Each dataset directory contains the following processed files:
  • graph.pkl: DGLGraph object for storing the graph structure.
  • split_edge.pkl: Splitted training samples, validation samples and test samples.
  • louvain_dataset.pkl: Detected community memberships through Louvain algorithm.
  • Other processed files for running PPRGo and UltraGCN, such as constrain_mat.pkl, ii_topk_neighbors.np.pkl.

Run the codes

All arguments are properly set in advance in the script files for reproducing our results.

Here we take GraphSAGE and GAT as examples.

> bash script/run_graphsage_e2e.sh
> bash script/run_gat_e2e.sh

BibTeX

If you like our work and use the model for your research, please cite our work as follows.

@inproceedings{luo2023cross-links,
author = {Luo, Zihan and Huang, Hong and Lian, Jianxun and Song, Xiran and Xie, Xing and Jin, Hai},
title = {Cross-links Matter for Link Prediction: Rethinking the Debiased GNN from a Data Perspective},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year = {2023},
month = {October},
url = {https://www.microsoft.com/en-us/research/publication/cross-links-matter-for-link-prediction-rethinking-the-debiased-gnn-from-a-data-perspective/},
}