Graph Contrastive Learning Meets Graph Meta Learning: A Unified Method for Few-shot Node Tasks
Hao Liu11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPT Jiarui Feng11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPT Lecheng Kong11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPT Dacheng Tao33{}^{3}start_FLOATSUPERSCRIPT 3 end_FLOATSUPERSCRIPT Yixin Chen11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPT Muhan Zhang22{}^{2}start_FLOATSUPERSCRIPT 2 end_FLOATSUPERSCRIPT {liuhao, feng.jiarui, jerry.kong, 11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPTWashington University in St. Louis 22{}^{2}start_FLOATSUPERSCRIPT 2 end_FLOATSUPERSCRIPTInstitute for Artificial Intelligence, Peking University 33{}^{3}start_FLOATSUPERSCRIPT 3 end_FLOATSUPERSCRIPTJD Explore Academy
Abstract
Graph Neural Networks (GNNs) have become popular in Graph Representation Learning (GRL). One fundamental application is few-shot node classification. Most existing methods follow the meta learning paradigm, showing the ability of fast generalization to few-shot tasks. However, recent works indicate that graph contrastive learning combined with fine-tuning can significantly outperform meta learning methods. Despite the empirical success, there is limited understanding of the reasons behind it. In our study, we first identify two crucial advantages of contrastive learning compared to meta learning, including (1) the comprehensive utilization of graph nodes and (2) the power of graph augmentations. To integrate the strength of both contrastive learning and meta learning on the few-shot node classification tasks, we introduce a new paradigm—Contrastive Few-Shot Node Classification (COLA). Specifically, COLA employs graph augmentations to identify semantically similar nodes, which enables the construction of meta-tasks without the need for label information. Therefore, COLA can utilize all nodes to construct meta-tasks, further reducing the risk of overfitting. Through extensive experim
中文速览
少样本节点分类（few-shot node classification）长期依赖元学习范式，但这类方法只能用带标签的训练类节点构建任务，既浪费了大量未标注节点的信息，又容易过拟合到训练类上。作者首先通过消融实验证明，图对比学习之所以能显著超越元学习，关键在于两点：它借助图数据增强学到了更具判别力的表示，且能直接利用全图所有节点（包括验证集和测试集节点）进行预训练，从而避免过拟合。基于这一洞察，他们提出了新范式 COLA（Contrastive Few-Shot Node Classification），核心思路是用三个增强视图的嵌入一致性来判断语义相似性，从而在完全不依赖标签的情况下构建元任务——随机采样查询节点，再从动量编码器生成的候选池中找出跨增强视图最相似的节点作为支持集，最后用监督对比损失训练整个框架。在六个真实世界数据集上的实验表明，COLA 在所有任务上均达到新的最优水平，验证了将对比学习的全节点利用与数据增强优势融入元学习框架的有效性。
原文 arXiv:2309.10376;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2309.10376v1