Skip to content

[CIKM 2024] Do We Really Need Graph Convolution During Training? Light Post-Training Graph-ODE for Efficient Recommendation

License

Notifications You must be signed in to change notification settings

DavidZWZ/LightGODE

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

28 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Do We Really Need Graph Convolution During Training? Light Post-Training Graph-ODE for Efficient Recommendation

Pytorch Implementation for CIKM 2024 Full Research Track Paper:

Do We Really Need Graph Convolution During Training? Light Post-Training Graph-ODE for Efficient Recommendation. 33rd ACM International Conference on Information and Knowledge , CIKM 2024
Weizhi Zhang, Liangwei Yang, Zihe Song, Henry Peng Zou, Ke Xu, Liancheng Fang, Philip S. Yu

Investigation: Do We Really Need Graph Convolution During Training?

Investigate the Role of Graph Convolution & Analyze it from BFS (GCN) vs DFS (MF) perspective

Investigation BFS vs DFS

Identify the Embedding Discrepancy Issue when further Increase the Depth of Post-Training Graph Convolution Layers

ED

Solution: Light Post-Training Graph-ODE for Efficient Recommendation.

Framework

Set up:

Dependencies

pip install -r requirements.txt

Dataset Preparation

Get the Amazon_Beauty, Amazon_Toys_and_Games, gowalla dataset under dataset folder

Please modify the file name and arrange them in the following structure:

LightGODE/
    ├─ Dataset/
        ├─ amazon-beauty
            ├─ amazon-beauty.inter
        ├─ amazon-toys-games
            ├─ amazon-toys-games.inter
        ├─ gowalla
            ├─ gowalla.inter

Running on different datasets:

Amazon-Beauty

python run_recbole.py -m LightGODE -d amazon-beauty -t 1.8 -g 0.5

Amazon-Toys-and-Games

python run_recbole.py -m LightGODE -d amazon-toys-games -t 0.8 -g 0.5 -w 0

Gowalla

python run_recbole.py -m LightGODE -d gowalla -t 3 -g 10 -w 0

Citation

If you find this repository helpful, please kindly consider citing our paper

@article{zhang2024we,
  title={Do We Really Need Graph Convolution During Training? Light Post-Training Graph-ODE for Efficient Recommendation},
  author={Zhang, Weizhi and Yang, Liangwei and Song, Zihe and Zou, Henry Peng and Xu, Ke and Fang, Liancheng and Yu, Philip S},
  journal={arXiv preprint arXiv:2407.18910},
  year={2024}
}

Acknowledgement

The structure of this repo is built based on RecBole. Thanks for their great work.

About

[CIKM 2024] Do We Really Need Graph Convolution During Training? Light Post-Training Graph-ODE for Efficient Recommendation

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages