L2CL: Embarrassingly Simple Layer-to-Layer Contrastive Learning for Graph Collaborative Filtering

Requirements

python>=3.9.18
pytorch>=1.13.1
torch-geometric>=2.4.0
torch-sparse>=0.6.17+pt113cu117
numpy>=1.26.1
pandas>=2.1.2
CUDA 11.7

Installation

conda create -n L2CL python=3.9
conda activate L2CL

pip install torch==1.13.1+cu117 -f https://download.pytorch.org/whl/torch/
pip install torch-sparse==0.6.17 -f https://pytorch-geometric.com/whl/torch-1.13.1+cu117.html
pip install torch-scatter==2.1.1 -f https://pytorch-geometric.com/whl/torch-1.13.1+cu117.html
pip install torch-geometric
pip install -r requirements.txt

Dataset

Datasets	#Users	#Items	#Interactions	Density
Kindle	60,468	57,212	880,859	0.00025
Yelp	45,477	30,708	1,777,765	0.00127
Books	58,144	58,051	2,517,437	0.00075
QB-video	30,323	25,730	1,581,136	0.00203

For amazon-kindle-store, yelp, amazon-books, they will be automatically downloaded via RecBole once you run the main program.

For QB-video, we provide it under dataset/

cd dataset
unzip QB-video.zip

Reproduction

We integrate our L2CL method into the RecBole and RecoBole-GNN framework.

You can reproduct our experiment results through below instructions:

Amazon-kindle-store

bash scripts/run_kindle.sh

Yelp

bash scripts/run_yelp.sh

Amazon-books

bash scripts/run_books.sh

QB-video

bash scripts/run_video.sh