基于模糊超图神经网络的节点分类方法
DOI:
作者:
作者单位:

华东交通大学理学院,江西 南昌 330013

作者简介:

熊婉悦(1999—),女,硕士研究生,研究方向为人工智能、深度学习。E-mail:531818418@qq.com。

通讯作者:

中图分类号:

TP183

基金项目:

国家自然科学基金项目(61991401);衢州市科技计划项目(2023K265)


Node Classification Method Based on Fuzzy Hypergraph Neural Network
Author:
Affiliation:

School of Science, East China Jiaotong University, Nanchang 330013 , China

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    目的】超图神经网络(HGNN)具有学习类间唯一性和类内共性的能力,可以显著提高学习性能。但是,传统HGNN方法缺乏决定低维数据节点间如何进行连接交互的强关系归纳。针对此问题,提出一种基于模糊理论的模糊HGNN(FHGNN) 节点分类算法,根据数据节点的特征信息构建超图结构,加强了图的节点信息对节点连接的影响。【方法】FHGNN首先采用了一个边聚焦的图神经网络(GNN),通过边标签的迭代更新进行边预测。并根据边预测的输出设计模糊隶属度函数,以实现更精确的节点间连接关系表示。最后通过得到的关系表示构造超图,并再次对节点进行分类训练得到结果。在FHGNN中使用了边标签损失函数和节点标签损失函数并分别对其参数进行训练学习。【结果】实验结果表明,所提的FHGNN方法更能够适应小规模低维数据,并在节点分类任务上取得好的效果。【结论】对于不同数据集的分类任务,FHGNN可以更有效学习节点的相关特征信息,提高学习的效果。

    Abstract:

    Objective】Hypergraph neural networks (HGNN) have the ability to learn inter-class uniqueness and intra-class commonality, which can significantly improve learning performance. However, traditional HGNN methods are in lack of the strong relational induction which determines the way how low-dimensional data nodes interact with each other. In order to solve this problem, a fuzzy HGNN(FHGNN) classification algorithm based on fuzzy theory is proposed, and hypergraph structure is constructed according to the characteristic information of data nodes.【Method】FHGNN first adopts an edge-focused GNN to make edge prediction through iterative updates of edge labels. The fuzzy membership function is designed according to the output of edge prediction to achieve a more accurate representation of the connection relationship between nodes. Finally, the hypergraph is constructed by the relation representation. Then the nodes are classified again and the result is obtained. The edge label loss function and node label loss function are used in FHGNN and their parameters are trained and learned respectively.【Result】Experimental results prove the proposed FHGNN method is more suitable for small-scale data with low node information dimension, and performs well in node classification tasks.【Conclusion】For classification tasks of different data sets, FHGNN can learn the relevant feature information of nodes more effectively and improve the learning effect.

    参考文献
    相似文献
    引证文献
引用本文

熊婉悦,徐峰,黄宇廷,韩星宇,范自柱.基于模糊超图神经网络的节点分类方法[J].华东交通大学学报,2024,41(3):119-126.
Xiong Wanyue, Xu Feng, Huang Yuting, Han Xingyu, Fan Zizhu. Node Classification Method Based on Fuzzy Hypergraph Neural Network[J]. JOURNAL OF EAST CHINA JIAOTONG UNIVERSTTY,2024,41(3):119-126

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2023-10-24
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2024-07-09
  • 出版日期: