文献类型：期刊文献

英文题名：Co-embedding of nodes and attributes with hypergraph neural networks

作者：Li, Jinsheng[1];Lu, Yisong[1];Wang, Zhineng[1];Wei, Pengfei[1];Zhang, Xueqin[1]

机构：[1]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China

年份：2026

卷号：680

外文期刊名：NEUROCOMPUTING

收录：;EI(收录号：20261120285709);WOS:【SCI-EXPANDED(收录号:WOS:001719411500001)】；

基金：This work was supported by the Major Program of National Fund of Philosophy and Social Sciences of China (grant number: 23&ZD142) .

语种：英文

外文关键词：Attribute networks; Network embedding; Bipartite graphs; Hypergraphs; Hypergraph neural networks

摘要：Effective node embeddings are crucial for analyzing attribute networks, and the fusion of node structures and attributes is key to enhancing embedding quality. However, existing attribute network embedding methods often suffer from poor embedding performance and low efficiency. To address these challenges, we propose a novel network node and attribute co-embedding method named CnaHGNN based on a hypergraph neural network. To enhance the expressive power of the graph model without significantly increasing the network's topological com plexity, bipartite graph and hypergraph structures are introduced in this method. At the same time, the attribute node embeddings from the bipartite graph are aggregated as node embeddings in the hypergraph network. A hypergraph neural network is then leveraged to extract high-order embeddings for hypergraph nodes. The node and attribute embeddings in the bipartite graph are iteratively updated through an unsupervised link prediction task. Additionally, a reconstruction function and an adaptive dynamic adjustment strategy are introduced dur ing training to optimize the aggregation process, improving both the quality and robustness of the embeddings. Empirical results demonstrate that CnaHGNN consistently outperforms state-of-the-art methods across various real-world applications, highlighting its effectiveness and generalizability.