Representation Learning of Temporal Graphs with Structural Roles

1. Environment

pip install -r requirements.txt

2. File Structure

• code:
  • Data-related codes:utils/minibatch.py,utils/preprocess.py,utils/random_walk.py,utils/utilities.py
  • Model-related codes:models/model.py
  • Training validation and testing-related :train.py
• data: Our data folder. data/DBLP3,data/DBLP5.

3. Quick Start

Dataset

Here, using DBLP3 and DBLP5 as examples, the following file structures are outlined: The .npz files contain original graph structure data. The _deg_nc files consist of role sets identified by a degree-based structural role discovery algorithm. The _role_nc files contain role sets derived from a motif-based structural role discovery algorithm, and the _wl_nc files hold role sets determined by a Weisfeiler-Lehman-based structural role discovery algorithm.

Training

For link prediction

python train.py --task==link prediction

For node classification

python train.py --task==node classification

4. Acknowledgements

• This code is based on the works available at DySAT (https://github.com/aravindsankar28/DySAT) and EvolveGCN (https://github.com/IBM/EvolveGCN).

• The code will be continuously improved and updated.

5. Citation

• If you find our work helpful, please consider citing the following.

@inproceedings{du2024representation,
  title={Representation Learning of Temporal Graphs with Structural Roles},
  author={Du, Huaming and Shi, Long and Chen, Xingyan and Zhao, Yu and Zhang, Hegui and Yang, Carl and Zhuang, Fuzhen and Kou, Gang},
  booktitle={Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
  pages={654--665},
  year={2024}
}