非同质性大型图数据集
收藏资源简介:
本研究收集了一系列大型非同质性图数据集,用于图机器学习评估。这些数据集涵盖了从在线社交网络到引文网络等多个应用领域,具有比以往数据集更多的节点和边。例如,Penn94数据集用于预测用户性别,而arXiv-year数据集则用于预测论文发表年份。这些数据集不仅规模更大,而且应用领域更广,能够捕捉到不同类型的复杂标签-拓扑关系。通过这些数据集,研究者们进行了广泛的实验,比较了14种图学习方法和3种图小批量技术,展示了LINKX模型在非同质性图学习中的优越性能。
This study collects a series of large-scale heterophilic graph datasets for graph machine learning evaluation. These datasets cover multiple application domains ranging from online social networks to citation networks, and contain more nodes and edges than previous datasets. For example, the Penn94 dataset is used for user gender prediction, while the arXiv-year dataset is designed for paper publication year prediction. These datasets not only have larger scales and wider application scopes, but also can capture complex label-topology relationships of various types. Using these datasets, researchers have conducted extensive experiments, compared 14 graph learning methods and 3 graph mini-batch techniques, and demonstrated the superior performance of the LINKX model in heterophilic graph learning.




