Tolokers-Fair, FB-Penn94-Fair, Pokec-Fair
收藏资源简介:
本文介绍了三个新的GNN公平性基准数据集:Tolokers-Fair、FB-Penn94-Fair和Pokec-Fair。这些数据集由密歇根大学的研究人员创建,旨在解决图神经网络(GNN)中的公平性问题。每个数据集都包含了多样化的局部同质性水平,以便研究在不同分布外(OOD)设置下的公平性问题。数据集的创建过程包括提取最大连通组件并确保数据集具有明确的敏感属性信息。这些数据集的应用领域主要集中在节点分类任务中的公平性评估,旨在解决因局部同质性差异导致的预测不公平问题。
This paper introduces three novel GNN fairness benchmark datasets: Tolokers-Fair, FB-Penn94-Fair, and Pokec-Fair. Developed by researchers from the University of Michigan, these datasets are designed to address fairness issues in Graph Neural Networks (GNNs). Each dataset features diverse levels of local homophily, enabling studies on fairness problems under various out-of-distribution (OOD) settings. The dataset creation process involves extracting the largest connected component and ensuring that the datasets have clear sensitive attribute information. The primary application of these datasets focuses on fairness evaluation for node classification tasks, aiming to solve prediction unfairness caused by disparities in local homophily.




