Text-space Graph Foundation Models

本数据集由密歇根州立大学创建，涵盖超过20个不同领域的数据集，包括学术、电子商务、生物学等。数据集通过将原始数据文件预处理或生成专家描述转换为文本空间，以支持全面的评估。这些数据集用于解决图机器学习中的特征异质性问题，通过统一的文本空间模型提高模型性能。数据集的应用领域广泛，旨在通过统一的模型框架解决多任务和多数据集的挑战，推动图基础模型的发展。

This dataset collection was developed by Michigan State University, encompassing datasets spanning over 20 distinct domains such as academia, e-commerce, biology, and more. All datasets are converted into a unified textual space through preprocessing raw data files or generating expert-written descriptions to enable comprehensive evaluations. These datasets are tailored to address the feature heterogeneity issue in graph machine learning, and enhance model performance by leveraging unified textual space models. With broad application prospects, this collection aims to tackle the challenges of multi-task and multi-dataset learning via a unified model framework, and advance the development of graph foundation models.

搜集汇总

背景与挑战

背景概述

该数据集包含多个图数据集，覆盖计算机科学引用、电子商务、生物引用和知识图谱等多个领域，支持节点分类、链接预测和图分类等任务。数据集提供了详细的图结构信息（如节点数、边数）和预处理文件，适用于图基础模型的训练和评估。