OPENGRAPHXAI
收藏资源简介:
OPENGRAPHXAI是一个基于现实世界分子图的XAI基准数据集,旨在解决图神经网络(GNNs)决策过程不透明的问题。该数据集由15个图分类数据集组成,每个数据集都包含已知的真实解释(GT)模型,用于评估解释的质量。数据集的大小、数据量和Tokens数等信息未在论文中明确提及,但提到数据集是基于11个现有的分子图数据集构建的,且可以通过代码生成超过2000个额外的XAI基准数据集。数据集的创建过程使用了Weisfeiler-Leman着色算法来识别与每个类别相关的子图模式。该数据集的应用领域是图XAI,旨在帮助科学家评估和改进图解释方法。
OPENGRAPHXAI is an XAI benchmark dataset built on real-world molecular graphs, developed to address the opaqueness issue in the decision-making processes of Graph Neural Networks (GNNs). This dataset comprises 15 graph classification datasets, each equipped with Ground Truth (GT) models that provide verified authentic explanations for evaluating the quality of generated explanations. Specific details including the dataset size, data volume, and token count are not explicitly specified in the accompanying paper. Nevertheless, it is noted that the dataset is constructed upon 11 existing molecular graph datasets, and over 2000 additional XAI benchmark datasets can be generated using the provided code. The dataset creation process adopts the Weisfeiler-Leman coloring algorithm to identify subgraph patterns correlated with each category. The application scope of this dataset lies in graph XAI, aiming to assist scientists in evaluating and refining graph explanation methods.




