Skip to main content
QUICK REVIEW

[论文解读] BAED: a New Paradigm for Few-shot Graph Learning with Explanation in the Loop

Chao Chen, Xujia Li|arXiv (Cornell University)|Mar 2, 2026
Advanced Graph Neural Networks被引用 0
一句话总结

BAED 引入一个解释闭环框架,用基于信念传播的标签增强、用于解释子图提取的辅助 GNN,以及基于梯度的解释,以在结构型或特征稀缺的图上提升准确性和可解释性。

ABSTRACT

The challenges of training and inference in few-shot environments persist in the area of graph representation learning. The quality and quantity of labels are often insufficient due to the extensive expert knowledge required to annotate graph data. In this context, Few-Shot Graph Learning (FSGL) approaches have been developed over the years. Through sophisticated neural architectures and customized training pipelines, these approaches enhance model adaptability to new label distributions. However, compromises in extcolor{black}{the model's} robustness and interpretability can result in overfitting to noise in labeled data and degraded performance. This paper introduces the first explanation-in-the-loop framework for the FSGL problem, called BAED. We novelly employ the belief propagation algorithm to facilitate label augmentation on graphs. Then, leveraging an auxiliary graph neural network and the gradient backpropagation method, our framework effectively extracts explanatory subgraphs surrounding target nodes. The final predictions are based on these informative subgraphs while mitigating the influence of redundant information from neighboring nodes. Extensive experiments on seven benchmark datasets demonstrate superior prediction accuracy, training efficiency, and explanation quality of BAED. As a pioneer, this work highlights the potential of the explanation-based research paradigm in FSGL.

研究动机与目标

  • 在少-shot 图学习中解决标签稀缺性和噪声问题,而不依赖丰富的节点特征。
  • 提出一个解释闭环管线,以提升 FSGL 的鲁棒性和可解释性。
  • 实现与多种 GNN 主干兼容的梯度基子图解释。
  • 通过辅助子图聚焦学习和 BP 驱动增强来提升训练效率。)

提出的方法

  • 基于信念传播的标签增强,将先验信息从标记节点传播到未标记节点。
  • 在先验上训练的辅助 GNN,识别包含决定性信息的解释性子图。
  • 使用反向传播对解释性子图进行边的排序,形成前 N 个节点的子图。
  • 通过再次应用 BP 对解释性子图进行决策,产生最终预测。
  • 端到端 BAED 框架,支持多种子图解释方法和主干网络。

实验结果

研究问题

  • RQ1BP 基标签增强能否在 FSGL 中提升标签质量并将信息传播到局部邻域以外?
  • RQ2在 augmented priors 上训练的辅助 GNN 是否能有效识别用于解释的信息子图?
  • RQ3对解释性子图的梯度基解释是否能在少-shot 设置中获得真实且鲁棒的预测?
  • RQ4BAED 流水线相较于 FSGL 基线在不同数据集和标注比例下的表现如何?

主要发现

DatasetSAGEGATGINGCNSGCDLRGAEHiD-NetGPNtsGCNDCISAGE+IGSAGE+SMBAEDImprove.
Cora0.3010.3070.2690.311*0.3010.3010.3000.2510.2960.2380.5150.5180.--66.5%
Citeseer0.2020.1920.215*0.1990.1890.1890.2040.1990.2110.1990.7690.7870.--266.1%
PubMed0.4000.3930.3920.3960.4100.3990.4840.4290.492*0.4590.4870.5070.--3.0%
Wiki0.1570.1520.1510.1520.1520.1520.161*0.1580.1520.1390.1720.1760.--9.3%
DBLP0.4430.4430.4430.4430.4480.4430.4470.517*0.4420.4490.6900.6980.--24.6%
Wisconsin0.2730.2690.2730.2690.3960.3800.2730.4070.2690.420*0.4410.4200.--4.9%
CoauthorCS0.2260.2160.1190.1830.280*0.1810.2780.2590.2550.2600.2100.1920.---25.1%
CoauthorPhy0.3080.3150.3080.3020.2890.2660.3200.352*0.3250.3110.5360.4860.--52.3%
  • 与最先进基线相比,BAED 在七个基准数据集上实现了更高的预测准确性、训练效率和解释质量。
  • 通过引入解释闭环学习,在特征无关的 FSGL 场景中获得显著提升。
  • 解释性子图比随机游走更频繁地富含标注信息,表明子图提取有效。
  • 可将梯度基解释与多种子图解释方法(如 IG、GNNExplainer 等)集成。
  • BAED 在不同 shot 数量的数据集上保持鲁棒性能,尽管在如 CoauthorCS 这类密集图上可能因 BP 的标签干扰而略有下降。

更好的研究,从现在开始

从论文设计到论文写作,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。