kirinmin / kgcl-sigir22 Goto Github PK

03-07-2023: Check our new implementation of KGCL based on the same code framework with KGIN at https://github.com/HKUDS/KGRec

17-01-2023: We have rebuilt the code for KGCL to significantly improve the readability and model performance! The new version will be available soon after close check.

This is the Pytorch implementation for our SIGIR'22 paper: Knowledge Graph Contrastive Learning for Recommendation. The CF learning part in the code is based on the open-source repository here: LightGCN, many thanks to the authors!

You are welcome to cite our paper:

@inproceedings{kgcl2022,
  author = {Yang, Yuhao and Huang, Chao and Xia, Lianghao and Li, Chenliang},
  title = {Knowledge Graph Contrastive Learning for Recommendation},
  year = {2022},
  booktitle = {Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval},
  pages = {1434–1443}
}

pip install -r requirements.txt

We provide three processed datasets and the corresponding knowledge graphs: Yelp2018 and Amazon-book and MIND.

run KGCL on Yelp2018 dataset:

• change base directory

Change ROOT_PATH in code/world.py

• command

cd code && python main.py --dataset=yelp2018

cd code && python main.py --dataset=amazon-book

cd code && python main.py --dataset=MIND

We also simply implement LightGCN (SIGIR'20) and SGL (SIGIR'21) for easy comparison. You can test these models implemented here by:

cd code && python main.py --dataset=yelp2018 --model=lgn

and

cd code && python main.py --dataset=yelp2018 --model=sgl

However, we still recommend to also refer to the authors' official implementation to avoid potential performance problems.