Prompt-Driven Continual Graph Learning

Qi Wang, Tianfei Zhou, Ye Yuan, Rui Mao

---

This is the official repository for the paper Prompt-Driven Continual Graph Learning.

The following figure illustrates the overall PromptCGL framework, which mitigates catastrophic forgetting via hierarchical prompting and personalized prompt generation.

---

Experiment Environment

Our experiments are conducted on Python 3.10 and CUDA 12.2, with the following package dependencies:

torch==1.13.1            # core deep learning framework
dgl==1.1.3               # graph neural network library
torch-geometric==2.4.0   # PyG for deploying GNNs
ogb==1.3.6               # for Arxiv and Products datasets
dgllife==0.3.2           # graph datasets support
progressbar2             # visualization of training progress

---

Usage

To run training and evaluation on CoraFull with PromptCGL:

python main.py --dataset-name=corafull --cgl-method=PromptCGL

---

Cite

If you find this repository useful, please cite:

@article{wang2025prompt,
  title={Prompt-Driven Continual Graph Learning},
  author={Wang, Qi and Zhou, Tianfei and Yuan, Ye and Mao, Rui},
  journal={arXiv preprint arXiv:2502.06327},
  year={2025}
}

---

Credit

This repository was developed based on the CGLB.