[1]韩继辉,石玉鹏,黄子奇,等.结构-特征协同防御的图神经网络[J].郑州大学学报(工学版),2027,48(XX):1-9.[doi:10.13705/j.issn.1671-6833.2026.04.010]
　HAN Jihui,SHI Yupeng,HUANG Ziqi,et al.A Graph Neural Network for Structure-Feature Collaborative Defense[J].Journal of Zhengzhou University (Engineering Science),2027,48(XX):1-9.[doi:10.13705/j.issn.1671-6833.2026.04.010]
点击复制

结构-特征协同防御的图神经网络()
分享到：

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
48
期数:
2027年XX
页码:
1-9
栏目:
出版日期:
2027-12-10

文章信息/Info

Title:
A Graph Neural Network for Structure-Feature Collaborative Defense
作者:
韩继辉1 ;  石玉鹏1 ;  黄子奇2 ;  张安琳3 ;  黄道颖1
1.郑州轻工业大学 计算机科学与技术学院,河南 郑州 450001;2.北方信息控制研究院集团有限公司,江苏 南京 211153;3.郑州轻工业大学 工程训练中心,河南 郑州 450001
Author(s):
HAN Jihui1 ;  SHI Yupeng1 ;  HUANG Ziqi2 ;  ZHANG Anlin3 ;  HUANG Daoying1
1. College of Computer Science and Technology, Zhengzhou University of Light Industry, Zhengzhou 450001 , China; 2 . North Information Control Research Academy Group Co. , Ltd. , Nanjing 211153, China; 3. Engineering Training Center, Zhengzhou University of Light Industry, Zhengzhou 450001, China
关键词:
图神经网络;  鲁棒性; 结构扰动; 特征扰动; 稀疏注意力
Keywords:
graph neural networks;  robustness;  structure perturbation;  feature perturbation;  sparse attention
分类号:
TP18；TN929.5
DOI:
10.13705/j.issn.1671-6833.2026.04.010
文献标志码:
A
摘要:
为解决图神经网络在复杂扰动环境下的节点表征退化问题，本文提出一种结构-特征协同防御的图神经网络 SFCoRobustGNN。在结构层面引入稀疏注意力机制，融合结构先验以动态抑制异常边；在特征层面结合通道门控机制与非线性特征混合模块（FeatureMixPro），增强模型对特征扰动的适应能力；通过对抗训练与多目标优化策略，实现双路径协同防御。在 Cora、Citeseer 等多个基准数据集上的实验表明：面对不同强度的结构扰动（5%~40%）与特征攻击（ε=0.01~0.10），所提方法优于主流基线方法，节点分类准确率明显提升。在 ogbn-products 大规模数据集上，即使面对 20% 扰动率的 MetaAttack 攻击，仍能保持 71.82% 的准确率，展现了良好的扩展性。消融实验验证了各模块的有效性及协同效应。所提方法有效抑制了复杂扰动下的性能衰减，并展现出良好的泛化性。
Abstract:
To address the degradation of node representations in graph neural networks under complex perturbation environments, a structure-feature collaborative defense graph neural network named SFCoRobustGNN was proposed. Structurally, a sparse attention mechanism that integrated structure priors to dynamically suppress anomalous edges was introduce. Feature-wise, a channel gating mechanism was combined with a nonlinear feature mixing module (FeatureMixPro) to enhance the model’s adaptability to feature perturbations. A collaborative dual-pathway defense was achieved through adversarial training and a multi-objective optimization strategy. Experiments on multiple benchmark datasets, including Cora and Citeseer, demonstrated that the proposed method outperformed most mainstream baseline metods under various intensities of structure perturbations (5%–40%) and feature attacks (ε=0.01–0.10), showing significant improvement in node classification accuracy. On the large-scale ogbn-products dataset, it maintained an accuracy of 71.82% even under a 20% MetaAttack structure perturbation, demonstrating its strong scalability. Ablation studies validated the effectiveness and synergistic effects of each module. The proposed method effectively mitigated performance degradation under complex perturbations and exhibited excellent generalization.

参考文献/References:

[1] Zügner D, Akbarnejad A, Günnemann S. Adversarial attacks on neural networks for graph data[C]//Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York: ACM, 2018: 2847-2856.
[2] Zhu Yanqiao, Xu Yichen, Yu Feng, et al. Deep graph contrastive representation learning[PP/OL]. V2. arXiv(2020-07-13)[2025-09-30]. https://arxiv.org/abs/2006.04131.
[3] Battaglia P W, Hamrick J B, Bapst V, et al. Relational inductive biases, deep learning, and graph networks[PP/OL]. V3. arXiv(2018-10-17)[2025-09-30]. https://arxiv.org/abs/1806. 01261.
[4] Zhang Xiang, Zitnik M. GNNGuard: defending graph neural networks against adversarial attacks[PP/OL]. V3. arXiv(2020-10-28)[2025-09-30]. https://arxiv.org/abs/2006. 08149.
[5] Chen Ting, Kornblith S, Norouzi M, et al. A simple framework for contrastive learning of visual representations[PP/OL]. V3. arXiv(2020-07-01)[2025-09-30]. https://arxiv.org/abs/2002. 05709.
[6] Jin Wei, Ma Yao, Liu Xiaorui, et al. Graph structure learning for robust graph neural networks[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York: ACM, 2020: 66-74.
[7] Li Gege, Ye Zhonglin, Cao Shujuan, et al. An unsupervised link prediction algorithm based on an approximate graph neural network framework[J]. Journal of Zhengzhou University (Engineering Science), 2024, 45(6): 75-82. [李格格, 冶忠林, 曹淑娟, 等. 一种近似图神经网络框架的无监督链路预测算法[J]. 郑州大学学报(工学版), 2024, 45(6): 75-82.]
[8] Veličković P, Fedus W, Hamilton W L, et al. Deep graph infomax[PP/OL]. V2. arXiv(2018-12-21)[2025-09-31]. https://arxiv.org/abs/1809. 10341.
[9] Fanseu Kamboua B, Zhang Lin, Ma Kaili, et al. GRACE: A general graph convolution framework for attributed graph clustering[J]. ACM Transactions on Knowledge Discovery from Data, 2023, 17(3): 1-31.
[10] Zhu Yanqiao, Xu Yichen, Yu Feng, et al. Graph contrastive learning with adaptive augmentation[C]//Proceedings of the Web Conference 2021. New York: ACM, 2021: 2069-2080.
[11] Thakoor S, Tallec C, Azar M G, et al. Large-scale representation learning on graphs via bootstrapping[PP/OL]. V3. arXiv(2023-02-20)[2025-09-30]. https://doi.org/10.48550/arXiv.2102. 06514.
[12] Liu Xiaorui, Jin Wei, Ma Yao, et al. Elastic graph neural networks[PP/OL]. V1. arXiv(2021-07-05)[2025-09-30]. https://arxiv.org/abs/2107. 06996.
[13] Li Kuan, Liu Yang, Ao Xiang, et al. Reliable representations make a stronger defender: unsupervised structure refinement for robust GNN[C]//Proceedings of the 28th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: ACM, 2022: 925-935.
[14] Veličković P, Cucurull G, Casanova A, et al. Graph attention networks[EB/OL]. V3. arXiv(2018-02-04)[2025-09-30]. https://arxiv.org/abs/1710. 10903.
[15] Egressy B, Von Niederhäusern L, Blanuša J, et al. Provably powerful graph neural networks for directed multi-graphs[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2024, 38(10): 11838-11846.
[16] Ahmed N, Rossi R A, Zhou Rong. Cora[DS/OL]. [2025-10-10]. http://networkrepository.com/cora.php.
[17] Ahmed N, Rossi R A, Zhou Rong. Citeseer[DS/OL]. [2025-10-10]. http://networkrepository.com/citeseer.php.
[18] Adamic L, Glance N. Polblogs[DS/OL]. [2025-10-10]. http://networkrepository.com/polblogs.php.
[19] Namata G M, London B, Getoor L, et al. PubMed diabetes dataset[DS/OL]. [2025-10-10]. https://linqs.org/datasets/#pubmed-diabetes.
[20] HU Weihua, Fey M, Zitnik M, et al. Dataset ogbn-products[DS/OL]. Open Graph Benchmark, 2020. https://ogb.stanford.edu/docs/nodeprop/#ogbn-products.
[21] Zügner D, Borchert O, Akbarnejad A, et al. Adversarial attacks on graph neural networks: perturbations and their patterns[J]. ACM Transactions on Knowledge Discovery from Data, 2020, 14(5): 1-31.
[22] Waniek M, Michalak T P, Wooldridge M J, et al. Hiding individuals and communities in a social network[J]. Nature Human Behaviour, 2018, 2(2): 139-147.
[23] Madry A, Makelov A, Schmidt L, et al. Towards deep learning models resistant to adversarial attacks[PP/OL]. V4. arXiv(2019-09-04)[2025-10-10]. https://doi.org/10.48550/arXiv.1706.06083.
[24] Goodfellow I J, Shlens J, Szegedy C. Explaining and harnessing adversarial examples[PP/OL]. V3. arXiv(2015-03-20)[2025-09-30]. https://arxiv.org/abs/1412.6572.
[25] Kipf T N, Welling M. Semi-supervised classification with graph convolutional networks[PP/OL]. V4. arXiv(2017-02-22)[2025-09-30]. https://arxiv.org/abs/1609. 02907.
[26] Zhu Dingyuan, Zhang Ziwei, Cui Peng, et al. Robust graph convolutional networks against adversarial attacks[C]//Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York: ACM, 2019: 1399-1407.

相似文献/References:

[1]路玲,孙新德..基于图像子块DCT系数对的盲检测数字水印[J].郑州大学学报(工学版),2010,31(02):108112.
　LU Ling,SUN Xinde.A Blind Detection Digital Watermarking Algorithm Based on Pairs ofBlocks DCT Coefficients[J].Journal of Zhengzhou University (Engineering Science),2010,31(XX):108112.
[2]王珊,谷城..基于归一化自相关语音筛选的鲁棒性说话人确认研究[J].郑州大学学报(工学版),2009,30(03):141.
　Wang Shan,GU Cheng.Confirmation Researching on Robustness Speaker based-on NormalizedAutocorrelation Voice Selection/div>[J].Journal of Zhengzhou University (Engineering Science),2009,30(XX):141.
[3]徐振启,卢洵,罗少鹏..基于时空二维混沌的自适应彩色图像水印算法[J].郑州大学学报(工学版),2007,28(02):84.[doi:10.3969/j.issn.1671-6833.2007.02.022]
　Xu Zhenqi,Lu Xun,Luo Shaopeng.Adaptive color image watermarking algorithm based on two-dimensional chaos in space and time[J].Journal of Zhengzhou University (Engineering Science),2007,28(XX):84.[doi:10.3969/j.issn.1671-6833.2007.02.022]
[4]谭秀湖,刘国枝,孙旭..基于最小均方差下的3D模型数字水印算法[J].郑州大学学报(工学版),2006,27(04):101.[doi:10.3969/j.issn.1671-6833.2006.04.025]
　Tan Xiu Lake,Liu Guozhi,Sun Xu.Digital watermarking algorithm for 3D models based on minimum mean square deviation [J].Journal of Zhengzhou University (Engineering Science),2006,27(XX):101.[doi:10.3969/j.issn.1671-6833.2006.04.025]
[5]吴振龙,李 林,刘艳红.基于全驱控制方法的农机路径跟踪控制[J].郑州大学学报(工学版),2025,46(04):24.[doi:10.13705/j.issn.1671-6833.2025.04.019]
　WU Zhenlong,LI Lin,LIU Yanhong.Path Tracking of Agricultural Machinery Based on Fully Actuated Control Approaches[J].Journal of Zhengzhou University (Engineering Science),2025,46(XX):24.[doi:10.13705/j.issn.1671-6833.2025.04.019]
[6]郑红,罗俞建,凌侃,等.多模态跨尺度特征融合的药物靶标亲和力预测[J].郑州大学学报(工学版),2026,47(XX):1.[doi:10.13705/j.issn.1671-6833.2026.04.001]
　ZHENG Hong,LUO Yujian,LING Kan,et al.Multimodal Cross-scale Feature Fusion for Drug-Target Affinity Prediction[J].Journal of Zhengzhou University (Engineering Science),2026,47(XX):1.[doi:10.13705/j.issn.1671-6833.2026.04.001]
[7]张淑芬,李涛,张镇博,等.一种抗拜占庭攻击的联邦学习鲁棒聚合算法[J].郑州大学学报(工学版),2027,48(XX):1.[doi:10.13705/j.issn.1671-6833.2026.04.012]
　ZHANG Shufen,LI Tao,ZHANG Zhenbo,et al.A Robust Aggregation Algorithm Defending Against Byzantine Attacks in Federated Learning[J].Journal of Zhengzhou University (Engineering Science),2027,48(XX):1.[doi:10.13705/j.issn.1671-6833.2026.04.012]

备注/Memo

备注/Memo:
收稿日期:2026-01-28;修订日期:2026-02-28
基金项目:河南省科技攻关项目(252102211089;232102210064)
作者简介:韩继辉(1987— ) ,男,河南周口人,郑州轻工业大学副教授,博士,主要从事复杂网络、图深度学习方向研究,E-mail:hanjihui@ zzuli. edu. cn。
通信作者:黄道颖(1967— ) ,男,河南信阳人,郑州轻工业大学教授,博士,主要从事分布式计算、智能算法方向研究,E-mail:dyhuang@ zzuli. edu. cn。
更新日期/Last Update: 2026-03-13