Skip to content

[WWW2024] The official code for paper "Distributionally Robust Graph-based Recommendation System"

License

Notifications You must be signed in to change notification settings

WANGBohaO-jpg/DR-GNN

Repository files navigation

Distributionally Robust Graph-based Recommendation System (DR-GNN)

Static Badge

This is the PyTorch implementation for our WWW 2024 paper (oral).

Bohao Wang, Jiawei Chen, Changdong Li, Sheng Zhou, Qihao Shi, Yang Gao, Yan Feng, Chun Chen, Can Wang 2024. Distributionally Robust Graph-based Recommendation System. arXiv link

Requirements

To install requirements:

pip install -r requirements.txt

Training & Evaluation

You can get the results in the paper by running the following code.

Gowalla

python main.py --model=lgn --enable_DRO=1 --aug_on --full_batch --ood popularity_shift --dataset='gowalla' --weight_decay 0.0001 --alpha 0.04 --tau 1  --aug_coefficient 0.1 --aug_ratio 0.2

Douban

python main.py --model=lgn --enable_DRO=1 --aug_on --full_batch --ood popularity_shift --dataset='douban' --weight_decay 1e-7 --alpha 0.005 --tau 0.1 --aug_coefficient 0.04 --aug_ratio 0.05

Amazon Book

python main.py --model=lgn --enable_DRO=1 --aug_on --full_batch --ood popularity_shift --dataset='amazon-book' --weight_decay 0.0001 --alpha 0.03 --tau 0.8 --aug_coefficient 0.1 --aug_ratio 0.1

Yelp2018

python main.py --model=lgn --enable_DRO=1 --aug_on --full_batch --ood popularity_shift --dataset='yelp2018' --weight_decay 0.0001 --alpha 0.07 --tau 1 --aug_coefficient 0.25 --aug_ratio 0.05

Citation

If you find the paper useful in your research, please consider citing:

@inproceedings{wang2024distributionally,
  title={Distributionally Robust Graph-based Recommendation System},
  author={Wang, Bohao and Chen, Jiawei and Li, Changdong and Zhou, Sheng and Shi, Qihao and Gao, Yang and Feng, Yan and Chen, Chun and Wang, Can},
  booktitle={Proceedings of the ACM on Web Conference 2024},
  pages={3777--3788},
  year={2024}
}

Acknowledgments

This project makes use of the following open source projects:

  • LightGCN-PyTorch: Some of the functionalities are inspired by this project. Thanks for their great work.

About

[WWW2024] The official code for paper "Distributionally Robust Graph-based Recommendation System"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published