Temporal Graph Benchmark (TGB)

Temporal Graph Benchmark (TGB) 是一个包含9个挑战性和多样化数据集的集合，用于对机器学习模型在时序图上的现实、可复现和稳健评估。TGB数据集规模大，持续时间跨度多年，涵盖社交、贸易、交易和交通网络等多个领域。数据集包括节点和边级预测任务，设计了基于实际用例的评估协议。TGB提供了一个自动化的机器学习管道，包括数据加载、实验设置和性能评估，旨在促进时序图学习的未来研究。

Temporal Graph Benchmark (TGB) is a collection of 9 challenging and diverse datasets designed for realistic, reproducible and robust evaluation of machine learning models on temporal graphs. TGB datasets are large-scale, span multiple years in duration, and cover multiple domains including social, trade, transaction and transportation networks. The datasets include node-level and edge-level prediction tasks, with evaluation protocols formulated based on real-world use cases. TGB provides an automated machine learning pipeline covering data loading, experiment setup and performance evaluation, aiming to advance future research in temporal graph learning.