RELBENCH
收藏资源简介:
RELBENCH是由斯坦福大学创建的一个公共基准数据集,用于评估图神经网络在关系数据库上的预测任务。该数据集涵盖了电子商务、社交平台、医疗和体育等多个领域,包含从74K到41M不等的实体数量,以及3到15个表和15到140个列的多样结构。数据集的创建过程包括将关系数据转换为图表示,并使用深度表格模型提取节点特征。RELBENCH旨在解决关系数据库中预测任务的自动化问题,通过提供标准化的数据库和任务,支持研究者开发和比较新的关系深度学习方法。
RELBENCH is a public benchmark dataset created by Stanford University, designed to evaluate graph neural networks on prediction tasks over relational databases. This dataset covers multiple domains including e-commerce, social platforms, healthcare, sports and other fields. It boasts diverse structural configurations, with entity counts ranging from 74K to 41M, alongside 3 to 15 tables and 15 to 140 columns. The development pipeline of RELBENCH involves converting relational data into graph-based representations and extracting node features using deep tabular models. RELBENCH aims to address the automation of prediction tasks in relational databases, providing standardized databases and benchmark tasks to support researchers in developing and comparing novel relational deep learning approaches.




