文献类型：期刊文献

英文题名：A Sparse Transformer-Enhanced Graph Convolutional Model for Robust Node Importance Ranking in Complex Networks

作者：Huang, Ru[1];Zhou, Shuo[1];Li, Pengfei[1];Yang, Kun[1];Chen, Zijian[2];He, Jianhua[3];Chu, Xiaoli[4];Zhou, Zhengbing[5];Zhai, Guangtao[2]

机构：[1]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China;[2]Shanghai Jiao Tong Univ, Inst Image Commun & Informat Proc, Shanghai 200240, Peoples R China;[3]Univ Essex, Sch Comp Sci & Elect Engn, Colchester CO4 3SQ, England;[4]Univ Sheffield, Dept Elect & Elect Engn, Sheffield S1 3JD, England;[5]Shanghai Jiao Tong Univ, Shanghai Peoples Hosp 6, Sch Med, Shanghai 200233, Peoples R China

年份：2026

卷号：13

起止页码：2350

外文期刊名：IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING

收录：;EI(收录号：20254019277066);WOS:【SCI-EXPANDED(收录号:WOS:001638055400020)】；

基金：This work was supported in part by the National Natural Science Foundation of China under Grant 62322114, in part by the Fundamental Research Funds for the Central Universities under Grant YG2023LC06, in part by the Natural Science Foundation of Shanghai under Grant 20ZR1413800, and in part by the Shanghai Key Laboratory Open Project under Grant STCSM 22DZ2229005.

语种：英文

外文关键词：Transformers; Complex networks; Accuracy; Adaptation models; Computational modeling; Transfer learning; Machine learning algorithms; Computational efficiency; Transportation; Scalability; Node importance ranking; Sparse Transformer; Graph neural networks

摘要：Identifying and ranking influential nodes in complex networks is critical for broad applications in social, biological, transportation, and other infrastructure systems. Traditional centrality-based and heuristic methods often struggle to balance computational efficiency with accuracy and fail to capture long-range dependencies, limiting their effectiveness in large-scale and heterogeneous networks. To address these limitations, we propose a novel Sparse Transformer-based Graph Convolutional Network (STGCN) for robust and efficient node importance ranking. The STGCN integrates a hybrid architecture that combines Sparse Transformer layers and Graph Convolutional Networks (GCNs) to jointly model local topological features and global structural dependencies. Specifically, the Sparse Transformer layers employ a stochastic anchor mechanism and masked attention to reduce computational complexity while preserving critical long-range interactions. Additionally, a transfer learning strategy is introduced, where the model is pre-trained on synthetic Barab & aacute;si-Albert networks and then transferred to real-world graphs, enhancing generalization across diverse network topologies. Extensive experiments conducted on 15 real-world datasets demonstrate that STGCN significantly outperforms state-of-the-art methods in ranking consistency, achieving an average Kendall's Tau correlation coefficient of 0.7832, near-perfect monotonicity index scores, and superior top-k node identification accuracy. The proposed framework provides a scalable and generalizable solution for identifying key nodes in complex networks, enhancing network resilience and optimizing information dissemination.