EVENT DOSSIER
On September 7, 2026, arXiv cs.LG published a research paper titled “A Comparative Study of Counterfactual Explainers for Graph Neural Networks Enabling Multiple Types of Graph Edit”. This study aimed to compare the effectiveness of various counterfactual explainers in graph neural networks, focusing on their performance and applicability in supporting multiple types of graph editing operations.
本研究对比了六种最先进的反事实解释器模型，涵盖二元及多分类图与节点分类任务。研究基于真实世界和合成数据集，采用多种定量与定性指标评估性能。尽管支持增删边的反事实解释器已出现，但现有方法在解释大小、覆盖率和质量之间常存在权衡，且缺乏通用高效的方法。该研究旨在识别各方法的优劣，以指导未来研究方向。