# Explainability in Graph Neural Networks: A Taxonomic Survey

**Authors:** Hao Yuan, Haiyang Yu, Shurui Gui, Shuiwang Ji  
**arXiv:** [2012.15445](https://arxiv.org/abs/2012.15445)  
**Date:** 2020-12-31  
**Summary page:** https://www.summarizepaper.com/en/paper/2012.15445v1/

## Key points

- The field of artificial intelligence is rapidly advancing through deep learning methods
- Lack of interpretability is a major drawback of deep learning models
- Researchers have been developing post hoc techniques for explainability
- Recent progress in explainability has been notable in images and texts
- Advancements in graph data analysis led to the development of graph neural networks (GNNs) and their explainability methods
- Lack of standardized approaches for evaluating GNN explainability
- Hao Yuan, Haiyang Yu, Shurui Gui, and Shuiwang Ji conducted a comprehensive survey on GNN explainability
- The survey provides a unified view of current GNN explainability methods
- A standardized testbed for evaluations and benchmark datasets were created to facilitate further methodological developments
- The survey examines the state-of-the-art in GNN explainability methods and sets the stage for future advancements

## Summary

The field of artificial intelligence is rapidly advancing and has shown impressive performance in various tasks through the use of deep learning methods. However, a major drawback of these models is their lack of interpretability. To address this limitation, researchers have been developing post hoc techniques to explain the predictions made by these models, giving rise to the field of explainability. Recent progress in explainability has been particularly notable in the domains of images and texts. In parallel, advancements in graph data analysis have led to the development of graph neural networks (GNNs) and their associated explainability methods. Despite this progress, there remains a lack of standardized approaches for evaluating GNN explainability. To bridge this gap, a group of researchers including Hao Yuan, Haiyang Yu, Shurui Gui, and Shuiwang Ji conducted a comprehensive survey on explainability in GNNs. Their work provides a unified and taxonomic view of current GNN explainability methods, shedding light on both commonalities and differences among existing techniques. By establishing a standardized testbed for evaluations and generating benchmark graph datasets specifically tailored for GNN explainability assessments, they aim to facilitate further methodological developments in this area. Overall, this survey offers a detailed examination of the state-of-the-art in GNN explainability methods and sets the stage for future advancements in enhancing the interpretability and transparency of deep learning models applied to graph data analysis.