Skip to main content
QUICK REVIEW

[论文解读] Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction

Eli Chien, Wei-Cheng Chang|arXiv (Cornell University)|Oct 29, 2021
Advanced Graph Neural Networks被引用 14
一句话总结

该论文提出GIANT,一种自监督框架,通过多尺度邻域预测利用图拓扑增强节点特征提取,采用eXtreme Multi-label Classification(XMC)形式化和XR-Transformers。在ogbn-papers100M上,其下游GNN性能提升最高达1.42%,MLP性能提升13.86%,GraphSAGE性能提升3.1%,在多种图基准测试中展现出良好的可扩展性和鲁棒性。

ABSTRACT

Learning on graphs has attracted significant attention in the learning community due to numerous real-world applications. In particular, graph neural networks (GNNs), which take numerical node features and graph structure as inputs, have been shown to achieve state-of-the-art performance on various graph-related learning tasks. Recent works exploring the correlation between numerical node features and graph structure via self-supervised learning have paved the way for further performance improvements of GNNs. However, methods used for extracting numerical node features from raw data are still graph-agnostic within standard GNN pipelines. This practice is sub-optimal as it prevents one from fully utilizing potential correlations between graph topology and node attributes. To mitigate this issue, we propose a new self-supervised learning framework, Graph Information Aided Node feature exTraction (GIANT). GIANT makes use of the eXtreme Multi-label Classification (XMC) formalism, which is crucial for fine-tuning the language model based on graph information, and scales to large datasets. We also provide a theoretical analysis that justifies the use of XMC over link prediction and motivates integrating XR-Transformers, a powerful method for solving XMC problems, into the GIANT framework. We demonstrate the superior performance of GIANT over the standard GNN pipeline on Open Graph Benchmark datasets: For example, we improve the accuracy of the top-ranked method GAMLP from $68.25\\%$ to $69.67\\%$, SGC from $63.29\\%$ to $66.10\\%$ and MLP from $47.24\\%$ to $61.10\\%$ on the ogbn-papers100M dataset by leveraging GIANT.

研究动机与目标

  • 解决标准GNN流程中因节点特征提取与图无关而导致的次优性能问题。
  • 将图拓扑信息整合到类似BERT的预训练语言模型中,以改善节点表征学习。
  • 开发一种可扩展的自监督框架,利用邻域预测作为预训练任务。
  • 从理论上证明在图引导的表征学习中,使用XMC优于链接预测。
  • 在大规模图基准测试(如ogbn-papers100M)上展示GIANT的有效性和可扩展性。

提出的方法

  • 提出一种新颖的自监督任务——多尺度邻域预测,将每个节点的邻域编码为二值多标签向量。
  • 将邻域预测任务映射到eXtreme Multi-label Classification(XMC)问题,从而可使用最先进的XMC求解器(如XR-Transformers)。
  • 利用来自原始节点特征(如TF-IDF)和图结构的PIFA(Pairwise Interaction Feature Aggregation)嵌入,指导分层聚类以构建邻域。
  • 使用XMC目标微调预训练的BERT模型,生成图感知的节点特征,随后作为下游GNN的输入。
  • 采用多尺度聚类策略,捕捉不同距离下的邻域信息,提升异构图中表征的质量。
  • 集成XR-Transformer模型,高效处理XMC固有的大规模多标签预测任务,确保在十亿节点图上的可扩展性。

实验结果

研究问题

  • RQ1基于图结构的自监督能否在节点特征提取方面超越与图无关的方法(如BERT或TF-IDF)?
  • RQ2在图引导的自监督节点表征学习中,XMC是否比链接预测更有效?
  • RQ3在特征学习中同时整合原始节点属性与图拓扑信息,对下游GNN性能有何影响?
  • RQ4所提出的GIANT框架能否在大规模图(如ogbn-papers100M)上保持性能增益的同时实现可扩展性?
  • RQ5分层聚类和PIFA嵌入在邻域预测质量与下游准确率方面分别起到何种贡献?

主要发现

  • 在ogbn-papers100M数据集上,GIANT-XRT将最先进方法GAMLP-RLU的准确率从68.25%提升至69.67%,相对提升1.42%。
  • 在ogbn-papers100M上,GIANT-XRT将MLP性能从47.24%提升至61.10%,相对提升13.86%,在某些情况下优于更复杂的GNN模型。
  • 在ogbn-papers100M上,GIANT-XRT将SGC准确率从63.29%提升至66.10%,将GraphSAGE准确率从65.73%提升至68.83%,表明在各类模型上均实现稳定增益。
  • 消融实验表明,结合TF-IDF特征与PIFA构建的图邻域可获得最佳性能,而NO TFIDF+NO PIFA组合表现最差。
  • 在ogbn-products上,使用链接预测微调的BERT(BERT+LP)性能低于GIANT,表明邻域预测在自监督中比链接预测更具鲁棒性。
  • GIANT-XRT具有高度可扩展性,在多个OGB基准测试(包括ogbn-arxiv和ogbn-products)上均实现一致性能增益,其中在ogbn-products上MLP性能最高提升18.76%。

更好的研究,从现在开始

从阅读论文到最终审阅,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。