Feature propagation for link prediction

The ogbl-collab and ogbl-citation2 are two datasets for link prediction. The challenge leaderboard can be checked at: https://ogb.stanford.edu/docs/leader_linkprop/. We apply feature propagation to solve this challenge and this repo contains our code submission. The technical report can be checked at ogb_report.pdf

Requirements

Install base packages: Python==3.6 Pytorch==1.7.1 pytorch_geometric==2.0.1 ogb==1.3.2

Results on OGB Challenges

Running the default code 10 times, here we present our results on the ogbl-collab and ogbl-citation2.

Method	ogbl-collab (Hits@50)	ogbl-citation2 (MRR)
PLNLP	0.7046 ± 0.0040	--
PLNLP + SIGN	0.7087 ± 0.0033	--
MLP	0.1991 ± 0.0170	0.2900 ± 0.0018
MLP + SIGN	0.2839 ± 0.0127	0.3224 ± 0.0017

Training Process for ogbl-collab

1. PLNLP as backbone

python plnlp_sign.py --data_name=ogbl-collab  --predictor=DOT --use_valedges_as_input=True --year=2010 --train_on_subgraph=True --epochs=800 --eval_last_best=True --dropout=0.3 --gnn_num_layers=1 --grad_clip_norm=1 --use_lr_decay=True --random_walk_augment=True --walk_length=10 --loss_func=WeightedHingeAUC --data_path=dataset

2. MLP as backbone

python mlp_collab_sign.py

Training Process for ogbl-citation2

python mlp_citation2_sign.py

Reference

• https://ogb.stanford.edu/
• https://github.com/snap-stanford/ogb
• https://github.com/zhitao-wang/PLNLP
• Pairwise Learning for Neural Link Prediction (https://arxiv.org/pdf/2112.02936.pdf)

Citation

If you find this work useful, please consider citing the technical report:

@article{yao2022ogb,
  title={Technical Report for OGB Link Property Prediction},
  author={Yao, Liang and Liu, Qiang and Cai, Hongyun and Ji, Shenggong and He, Feng and Cheng, Xu},
  year={2022}
}