Semantic-Similarity Attention Meets Hypergraph Convolution for Scientific Publication Representation Learning
Ang LI , Yawen LI , Feifei KOU , Zhe XUE , Meiyu LIANG , Baoxiang WANG
Scientific publication data contain a wealth of valuable information, and numerous studies have proposed algorithms to mine such data. However, most existing non-hypergraph learning methods focus primarily on binary rela-tions in technical scientific publication data, while overlook-ing higher-order information embedded in non-binary inter-actions. Moreover, many hypergraph learning methods are unable to fully exploit both the structural characteristics of hypergraphs and the rich textual semantics present in sci-entific publication data. To address these limitations, we propose a hypergraph learning algorithm for scientific pub-lication data based on cross-channel hypergraph convolution and semantic-similarity attention, HCTA. Specifically, HCTA leverages cross-channel hypergraph convolution to capture higher-order structural dependencies and incorporates a text-semantic-similarity-based hypergraph attention mechanism to enhance representation learning and feature fusion. Exper-imental results on scientific publication PStyleification and sci-entific publication relationship prediction tasks demonstrate that our proposed HCTA achieves superior performance com-pared with the state-of-the-art representative methods.
Hypergraph learning / Hypergraph convolu-tional networks / Semantic-similarity attention / Scientific publication mining
Higher Education Press 2026
/
| 〈 |
|
〉 |
PDF
(6816KB)
