Does GCL Need a Large Number of Negative Samples? Enhancing Graph Contrastive Learning with Effective and Efficient Negative Sampling (AAAI 2025)

The official source code for "Does GCL Need a Large Number of Negative Samples? Enhancing Graph Contrastive Learning with Effective and Efficient Negative Sampling".

Part of code is referenced from Deep Graph Contrastive Representation Learning and PyGCL: A PyTorch Library for Graph Contrastive Learning)

Environment Setup

• torch==2.1.0
• torch-geometric==2.5.3
• torch-scatter==2.1.2
• torch-sparse==0.6.18
• scikit-learn==1.2.0
• scipy==1.10.1
• numpy==1.24.3

Usage

To train and test the model on a specific dataset, use:

bash run.sh <Dataset>  # Example: PubMed, CS, Photo, Computers, Physics, or Wiki-CS.