IGL-Bench

IGL-Bench是一个针对不平衡图学习的基础性综合基准，包含16个多样化的图数据集和24种不同的不平衡图学习算法。该数据集由北京航空航天大学创建，旨在解决图数据分布不平衡问题，特别是在节点级和图级任务中的类别不平衡和拓扑不平衡。数据集涵盖了从强同质性到强异质性的多种图数据，如引文网络、亚马逊共现网络等。创建过程中，采用了统一的数据处理和分割策略，以确保算法间的公平比较。该数据集的应用领域广泛，包括社交网络分析、生物信息学等，旨在通过提供一个公平的评估平台，推动不平衡图学习领域的研究进展。

IGL-Bench is a fundamental and comprehensive benchmark for imbalanced graph learning, which includes 16 diverse graph datasets and 24 distinct imbalanced graph learning algorithms. Developed by Beihang University, this benchmark aims to address the issue of imbalanced graph data distributions, particularly class imbalance and topological imbalance in both node-level and graph-level tasks. It covers a wide range of graph data spanning from strong homophily to strong heterophily, such as citation networks, Amazon co-occurrence networks and other typical graph datasets. During its construction, a unified data processing and splitting strategy was adopted to ensure fair comparisons between different algorithms. This benchmark has broad application scenarios including social network analysis, bioinformatics and more, with the core goal of providing a fair evaluation platform to advance research progress in the field of imbalanced graph learning.

搜集汇总

背景与挑战

背景概述

IGL-Bench是一个针对不平衡图学习的基础性基准，包含16个多样化图数据集和24种算法，采用统一处理标准确保公平比较，旨在推动社交网络分析、生物信息学等领域的研究进展。