This Project is for our work, $Fidelity_\alpha$ , TOWARDS ROBUST FIDELITY FOR EVALUATING EXPLAINABILITY OF GRAPH NEURAL NETWORKS(ICLR2024)[TrustLOG@WWW]

use fidelity as metric

• If you just want to use our fidelity for evaluation, please use tools folder.
• Please refer to example.py

for results reproduce

• generate samples, please run generate_edit_distance.py
• generate ori fidelity results, please run experiment_editdistance_ori_fid.py
• generate our fidelity results, please run experiment_editdistance_new_fid.py

Probability ori. Fidelity results of Ba2Motifs dataset(ACC results can be found in ./pictures), the x-axis means adding non-explanation edges to GT, y-axis means remove edges from GT. The following three figures are Original $Fidelity+$, $Fidelity-$, $Fidelity_\Delta$.

Probability our Fidelity results of Ba2Motifs dataset( $\alpha_1$ = 0.1, $\alpha_2$ = 0.9 )(ACC results can be found in ./pictures). The following three figures are Ours $Fidelity+$, $Fidelity-$, $Fidelity_\Delta$.

Probability ori. Fidelity results of TreeCycles dataset(ACC results can be found in ./pictures), the x-axis means adding non-explanation edges to GT, y-axis means remove edges from GT. The following three figures are Original $Fidelity+$, $Fidelity-$, $Fidelity_\Delta$.

Probability our Fidelity results of TreeCycles dataset( $\alpha_1$ = 0.1, $\alpha_2$ = 0.9 )(ACC results can be found in ./pictures). The following three figures are Ours $Fidelity+$, $Fidelity-$, $Fidelity_\Delta$.

Acknowledge. This project is base on [RE]-PGExplainer link

If this work is helpful for you, please consider citing our paper.

@article{zheng2023towards,
  title={Towards robust fidelity for evaluating explainability of graph neural networks},
  author={Zheng, Xu and Shirani, Farhad and Wang, Tianchun and Cheng, Wei and Chen, Zhuomin and Chen, Haifeng and Wei, Hua and Luo, Dongsheng},
  journal={arXiv preprint arXiv:2310.01820},
  year={2023}
}