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 of Role of Graph Convolution

BFS (GCN) vs DFS (MF) perspective

The Embedding Discrepancy Issue

pip install -r requirements.txt

Get the beauty, toys-and-games, gowalla dataset under dataset folder Overall file structure:

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

Amazon-Beauty

python run_recbole.py -m LightGODE -d amazon-beauty

Amazon-Toys-and-Games

python run_recbole.py -m LightGODE -d amazon-toys-games

Gowalla

python run_recbole.py -m LightGODE -d gowalla -w 0

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