This website requires JavaScript.

Beyond Real-world Benchmark Datasets: An Empirical Study of Node Classification with GNNs

Seiji MaekawaKoki NodaYuya SasakiMakoto Onizuka
Jun 2022
摘要
Graph Neural Networks (GNNs) have achieved great success on a nodeclassification task. Despite the broad interest in developing and evaluatingGNNs, they have been assessed with limited benchmark datasets. As a result, theexisting evaluation of GNNs lacks fine-grained analysis from variouscharacteristics of graphs. Motivated by this, we conduct extensive experimentswith a synthetic graph generator that can generate graphs having controlledcharacteristics for fine-grained analysis. Our empirical studies clarify thestrengths and weaknesses of GNNs from four major characteristics of real-worldgraphs with class labels of nodes, i.e., 1) class size distributions (balancedvs. imbalanced), 2) edge connection proportions between classes (homophilic vs.heterophilic), 3) attribute values (biased vs. random), and 4) graph sizes(small vs. large). In addition, to foster future research on GNNs, we publiclyrelease our codebase that allows users to evaluate various GNNs with variousgraphs. We hope this work offers interesting insights for future research.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答