On Data-Aware Global Explainability of Graph Neural Networks

被引:2
作者
Lv, Ge [1 ]
Chen, Lei [1 ,2 ]
机构
[1] HKUST, Hong Kong, Peoples R China
[2] HKUST GZ, Guangzhou, Peoples R China
来源
PROCEEDINGS OF THE VLDB ENDOWMENT | 2023年 / 16卷 / 11期
基金
美国国家科学基金会;
关键词
D O I
10.14778/3611479.3611538
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Graph Neural Networks (GNNs) have significantly boosted the performance of many graph-based applications, yet they serve as black-box models. To understand how GNNs make decisions, explainability techniques have been extensively studied. While the majority of existing methods focus on local explainability, we propose DAG-Explainer in this work aiming for global explainability. Specifically, we observe three properties of superior explanations for a pretrained GNN: they should be highly recognized by the model, compliant with the data distribution and discriminative among all the classes. The first property entails an explanation to be faithful to the model, as the other two require the explanation to be convincing regarding the data distribution. Guided by these properties, we design metrics to quantify the quality of each single explanation and formulate the problem of finding data-aware global explanations for a pretrained GNN as an optimizing problem. We prove that the problem is NP-hard and adopt a randomized greedy algorithm to find a near optimal solution. Furthermore, we derive an improved bound of the approximation algorithm in our problem over the state-of-the-art (SOTA) best. Experimental results show that DAG-Explainer can efficiently produce meaningful and trustworthy explanations while preserving comparable quantitative evaluation results to the SOTA methods.
引用
收藏
页码:3447 / 3460
页数:14
相关论文
共 71 条
[31]  
Lin Wanyu, 2021, P MACHINE LEARNING R, V139
[32]   Local Community Detection in Multiple Networks [J].
Luo, Dongsheng ;
Bian, Yuchen ;
Yan, Yaowei ;
Liu, Xiao ;
Huan, Jun ;
Zhang, Xiang .
KDD '20: PROCEEDINGS OF THE 26TH ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY & DATA MINING, 2020, :266-274
[33]  
Luo Dongsheng, 2020, ADVANCESINNEURALINFO
[34]   On Glocal Explainability of Graph Neural Networks [J].
Lv, Ge ;
Chen, Lei ;
Cao, Caleb Chen .
DATABASE SYSTEMS FOR ADVANCED APPLICATIONS, DASFAA 2022, PT I, 2022, :648-664
[35]  
Meliou A, 2011, Arxiv, DOI arXiv:1009.2021
[36]   A Clustering-Based Approach to the Explanation of Database Query Answers [J].
Moreau, Aurelien ;
Pivert, Olivier ;
Smits, Gregory .
FLEXIBLE QUERY ANSWERING SYSTEMS 2015, 2016, 400 :307-319
[37]  
Morris C, 2020, Arxiv, DOI [arXiv:2007.08663, 10.48550/arXiv.2007.08663]
[38]   Safe Pattern Pruning: An Efficient Approach for Predictive Pattern Mining [J].
Nakagawa, Kazuya ;
Suzumura, Shinya ;
Karasuyama, Masayuki ;
Tsuda, Koji ;
Takeuchi, Ichiro .
KDD'16: PROCEEDINGS OF THE 22ND ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING, 2016, :1785-1794
[39]   Temporal Graph Kernels for Classifying Dissemination Processes [J].
Oettershagen, Lutz ;
Kriege, Nils M. ;
Morris, Christopher ;
Mutzel, Petra .
PROCEEDINGS OF THE 2020 SIAM INTERNATIONAL CONFERENCE ON DATA MINING (SDM), 2020, :496-504
[40]  
Olah C, 2017, Distill, DOI DOI 10.23915/DISTILL.00007