Scholarlink

<p>Hypergraphs, as a form of high-order complex graphs, excel in expressing intricate relationships within non-Euclidean, infinite-dimensional spaces. However, conventional hypergraphs are more vulnerable to the heterophily problem for semi-supervised node classification tasks compared to low-order graphs, and even multilayer perceptrons. This paper first explores the underlying reasons for this phenomenon and summarizes the theoretical results corresponding to the heterophily problem. Based on these findings, a structural influence-based dynamic hypergraph neural network (SID-HGNN) framework is proposed. SID-HGNN combines a label influence-based dynamic structure with integrated structural and locational features to enhance the hypergraph framework’s ability to address the heterophily problem. The efficiency and superiority of SID-HGNN are validated through experimental comparison with 16 relevant methods on publicly available and self-collected fixed-wing unmanned aerial vehicle fault datasets.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Structural influence-based dynamic hypergraph for semi-supervised classification

  • 期刊论文

  • 发表日期：

    2026年5月13日

查看全文
  • Shaojun Liang,
  • Ying Zheng,
  • Yi Yang,
  • Zhiwei Wang,
  • Housheng Su

摘要

Hypergraphs, as a form of high-order complex graphs, excel in expressing intricate relationships within non-Euclidean, infinite-dimensional spaces. However, conventional hypergraphs are more vulnerable to the heterophily problem for semi-supervised node classification tasks compared to low-order graphs, and even multilayer perceptrons. This paper first explores the underlying reasons for this phenomenon and summarizes the theoretical results corresponding to the heterophily problem. Based on these findings, a structural influence-based dynamic hypergraph neural network (SID-HGNN) framework is proposed. SID-HGNN combines a label influence-based dynamic structure with integrated structural and locational features to enhance the hypergraph framework’s ability to address the heterophily problem. The efficiency and superiority of SID-HGNN are validated through experimental comparison with 16 relevant methods on publicly available and self-collected fixed-wing unmanned aerial vehicle fault datasets.