TempME: Towards the Explainability of Temporal Graph Neural Networks via Motif Discovery

This is the Pytorch Implementation of TempME:Towards the Explainability of Temporal Graph Neural Networks via Motif Discovery [arXiv]

Train a Base Model

To start, you'll need to train a base model. Our framework supports several base model types, including TGAT, TGN, and GraphMixer. To train your model, use the following command, replacing ${type} with your chosen base model type (e.g., tgat, tgn, graphmixer) and ${dataset} with the name of your dataset.

python learn_base.py --base_type ${type} --data ${dataset}

Train an Explainer

Once you have a base model, the next step is to train an explainer. Use the following command to train your explainer:

python temp_exp_main.py --base_type ${type} --data ${dataset}

Verify Enhancement Effect

To evaluate the effectiveness of the explanatory subgraphs extracted by the explainer, use the following command:

python enhance_main.py --data ${dataset} --base_type ${type}

Citation

If you find this work useful, please consider citing:

@article{chen2024tempme,
  title={Tempme: Towards the explainability of temporal graph neural networks via motif discovery},
  author={Chen, Jialin and Ying, Rex},
  journal={Advances in Neural Information Processing Systems},
  volume={36},
  year={2024}
}