您的位置：山东大学 -> 科技期刊社 -> 《山东大学学报（工学版）》

山东大学学报 (工学版) ›› 2026, Vol. 56 ›› Issue (3): 137-143.doi: 10.6040/j.issn.1672-3961.0.2025.034

• 机器学习与数据挖掘 • 上一篇    下一篇

基于多层级核心聚合GNN的图节点分类算法

唐凯1,王芳1*,刘建霞2   

  1. 1.太原理工大学电气与动力工程学院, 山西 太原 030024;2.太原理工大学电子信息工程学院, 山西 太原 030024
  • 发布日期:2026-06-09
  • 作者简介:唐凯(1999— ),男,河南驻马店人,硕士研究生,主要研究方向为深度学习. E-mail:tangkai_gm@163.com. *通信作者简介:王芳(1976— ),女,四川泸州人,副教授,硕士生导师,博士,主要研究方向为人工智能、机器视觉. E-mail:wangfang@tyut.edu.cn
  • 基金资助:
    山西省重点研发计划资助项目(202202130501016)

Graph node classification algorithm based on multi-level core aggregation GNN

TANG Kai1, WANG Fang1*, LIU Jianxia2   

  1. TANG Kai1, WANG Fang1*, LIU Jianxia2(1. School of Electrical and Power Engineering, Taiyuan University of Technology, Taiyuan 030024, Shanxi, China;
    2. School of Electronic Information Engineering, Taiyuan University of Technology, Taiyuan 030024, Shanxi, China
  • Published:2026-06-09

PDF (PC)

5

摘要： 针对节点分类任务中邻居结构信息利用不充分、多层传递节点特征模糊化的问题,提出一种多层级核心聚合图神经网络(multi-level core aggregation graph neural network, MCAG)模型。MCAG模型将图数据划分为双节点和多节点连通分量,自定义核心节点筛选机制构建总核心节点组,进而扩展出核心节点子图,组合成核心节点聚合层,并与GraphSAGE层、群组归一化层融合得到节点最终表示。试验结果表明,MCAG模型在Cora等6个数据集上节点分类准确率平均提升3.28%,在Amap数据集上表现与基线模型相当,整体性能稳定;训练时间较原始架构平均缩减50%,核心节点集采样方法性能优于随机游走采样,验证了MCAG模型的有效性和优越性。

关键词: 多层级核心聚合, 筛选机制, 核心节点, 节点分类, 图神经网络

Abstract: To address the problems of insufficient utilization of neighbor structure information and feature blurring caused by multi-layer propagation in node classification tasks, a multi-level core aggregation graph neural network(MCAG)model was proposed. The MCAG model divided the graph data into two-node and multi-node connected components. A custom core node selection mechanism was used to construct a total core node set, from which a core node subgraph was then extended. These subgraphs were combined to form a core node aggregation layer, which was fused with a GraphSAGE layer and a group normalization layer to obtain the final node representations. Experimental results showed that the MCAG model improved the node classification accuracy by an average of 3.28% on six datasets, including Cora. On the Amap dataset, the model performed comparably to baseline models, demonstrated stable overall performance.The training time was reduced by an average of 50% compared to the original architecture, and the performance of the core node set sampling method was superior to that of random walk sampling. These findings verified the effectiveness and superiority of the MCAG model.

Key words: multi-level core aggregation, selection mechanism, core node, node classification, graph neural network

中图分类号: 

  • TP391
[1] SANKAR A, LIU Y, YU J, et al. Graphneural networks for friend ranking in large-scale social platforms[C] //Proceedings of the Web Conference 2021.New York, USA: ACM, 2021: 2535-2546.
[2] MALLA A M, BANKA A A. A systematic review of deep graph neural networks: challenges, classification, architectures, applications & potential utility in bioinformatics[EB/OL].(2023-11-03)[2025-03-15]. https://arxiv.org/abs/2311.02127
[3] TALLAPALLY D, WANG J, POTIKA K, et al. Using-graph neural networks for social recommendations[J]. Algorithms, 2023, 16(11): 515.
[4] SCARSELLI F, GORI M, TSOI A C, et al. Thegraph neural network model[J]. IEEE Transactions on Neural Networks, 2009, 20(1): 61-80.
[5] WU Z H, PAN S R, CHEN F W, et al. A comprehensive survey on graph neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(1): 4-24.
[6] YE Z X, FENG Z Y, XIAO J M, et al. Heterogeneous graph neural network-based software developer recommendation[C] //Proceedings of the 18th EAI International Conference on Collaborative Computing: Networking, Applications and Worksharing. Hangzhou, China: Springer, 2022: 433-452.
[7] CHEN Y H, TANG X, QI X B, et al. Learning graph normalization for graph neural networks[J]. Neurocomputing, 2022, 493: 613-625.
[8] ITOH T D, KUBO T, IKEDA K. Multi-level attention pooling for graph neural networks: unifying graph representations with multiple localities[J]. Neural Networks, 2022, 145: 356-373.
[9] OISHI Y, KANEIWA K. Multi-duplicated characterization of graph structures using information gain ratio for graph neural networks[J]. IEEE Access, 2023, 11: 34421-34430.
[10] DUAN W, XUAN J Y, QIAO M Y, et al. Graph convolutional neural networks with diverse negative samples via decomposed determinant point processes[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 35(12): 18160-18171.
[11] GE Q Q, ZHAO Z Y, LIU Y D, et al. PSP: pre-training and structure prompt tuning for graph neural networks[C] //Proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases. Vilnius, Lithuania: Springer, 2024: 423-439.
[12] ROTH A, LIEBIG T. Rankcollapse causes over-smoothing and over-correlation in graph neural networks[EB/OL].(2023-08-31)[2025-03-15]. https://arxiv.org/abs/2308.16800
[13] CHEN R, ZHANG S, LEONG H U, et al. Redundancy-free message passing for graph neural networks[C] //Proceedings of the 36th International Conference on Neural Information Processing Systems. New Orleans, USA: ACM, 2022:4316-4327.
[14] WANG Y, HU L, CAO X F, et al. Enhancing locally adaptive smoothing of graph neural networks via laplacian node disagreement[J]. IEEE Transactions on Knowledge and Data Engineering, 2024, 36(3): 1099-1112.
[15] LI J H, FAN S H. GRNN: graph-retraining neural network for semi-supervised node classification[J]. Algorithms, 2023, 16(3): 126.
[16] ZHOU K X, HUANG X, LI Y N, et al. Towards deeper graph neural networks with differentiable group normalization[EB/OL].(2020-06-12)[2025-03-15]. https://arxiv.org/abs/2006.06972
[17] 武凯丽, 陈京荣. 基于节点重要性排序的局部社区检测算法[J]. 山东大学学报(工学版), 2025, 55(1): 77-85. WU Kaili, CHEN Jingrong. Local community detection algorithm based on the node importance ranking[J]. Journal of Shandong University(Engineering Science), 2025, 55(1): 77-85.
[18] GUO H M, WANG S L, YAN X F, et al. Node importance evaluation method of complex network based on the fusion gravity model[J]. Chaos, Solitons & Fractals, 2024, 183: 114924.
[19] WU J, QIU T, CHEN G. A general deep-learning approach to node importance identification[J]. Chaos, Solitons & Fractals, 2024, 188: 115501.
[20] TARJAN R E, ZWICK U. Finding strongcomponents using depth-first search[J]. European Journal of Combinatorics, 2024, 119: 103815.
[21] WEI L N, HE Z, ZHAO H, et al. Search to capture long-range dependency with stacking GNNs for graph classification[C] // Proceedings of the ACM Web Conference 2023. Austin, USA: ACM, 2023: 588-598.
[22] SEN P, NAMATA G, BILGIC M, et al. Collective classification in network data[J]. AI Magazine, 2008, 29(3): 93-106.
[23] NAMATA G, LONDON B, GETOOR L, et al. Query-driven active surveying for collective classification[C] // Proceedings of the 10th Workshop on Mining and Learning with Graphs. Edinburgh, UK: ACM, 2012: 1-8.
[24] LIU Y, TU W X, ZHOU S H, et al. Deep graph clustering via dual correlation reduction[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2022, 36(7): 7603-7611.
[25] LIU Y, YANG X H, ZHOU S H, et al. Simple contrastive graph clustering[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 35(10): 13789-13800.
[26] VASHISTHA S, KUMAR N. Focus where it matters: graph selective state focused attention networks[EB/OL].(2024-10-21)[2025-03-15]. https://arxiv.org/abs/2410.15849
[27] KIPF T N, WELLING M. Semi-supervised classification with graph convolutional networks[EB/OL].(2017-02-22)[2025-03-15]. https://arxiv.org/abs/1609.02907
[28] CHIEN E, PENG J H, LI P, et al. Adaptive universal generalized pageRank graph neural network[EB/OL].(2020-06-14)[2025-03-15]. https://arxiv.org/abs/2006.07988
[29] DENG Z J, DONG Y P, ZHU J. Batch virtual adversarial training for graph convolutional networks[J]. AI Open, 2023, 4: 73-79.
[30] RONG Y, HUANG W B, XU T Y, et al. DropEdge: towards deep graph convolutional networks on node classification[EB/OL].(2019-07-25)[2025-03-15]. https://arxiv.org/abs/1907.10903
[31] ZHOU K Q, DONG Y F,WANG K X,et al. Understanding and resolving performance degradation in deep graph convolutional networks[C] //Proceedings of the 30th ACM International Conference on Information & Knowledge Management. Queensland, Australia: ACM, 2021: 2728-2737.
[32] VELICKOVIC P, CUCURULL G, CASANOVA A, et al. Graph attention networks[EB/OL].(2018-02-04)[2025-03-15]. https://arxiv.org/abs/1710.10903
[33] GASTEIGER J, BOJCHEVSKI A, GÜNNEMANN S, et al. Predict then propagate: graph neural networks meet personalized pageRank[EB/OL].(2018-10-14)[2025-03-15]. https://arxiv.org/abs/1810.05997
[34] XU K, HU W, LESKOVEC J, et al. How powerful are graph neural networks[EB/OL].(2019-02-22)[2025-03-15]. https://arxiv.org/abs/1810.00826
[35] SHANNON C E. A mathematical theory of communi-cation[J]. The Bell System Technical Journal, 1948, 27(3): 390-393.
[1] 陈宇,孟广婷,宗臣,袁卫华,王洁宁,王星. 双侧协同过滤多模态推荐对比表示提升算法[J]. 山东大学学报 (工学版), 2026, 56(3): 106-117.
[2] 黎俊亮,蒋沅,吴珑雪,刘宇. 融合节点与边特征的增强Graph Transformer[J]. 山东大学学报 (工学版), 2026, 56(3): 118-126.
[3] 邓彬, 张宗包, 赵文猛, 罗新航, 吴秋伟. 基于云边协同和图神经网络的电动汽车充电站负荷预测方法[J]. 山东大学学报 (工学版), 2025, 55(5): 62-69.
[4] 武凯丽,陈京荣. 基于节点重要性排序的局部社区检测算法[J]. 山东大学学报 (工学版), 2025, 55(1): 77-85.
[5] 林振宇,邵蓥侠. 基于盖根堡多项式最佳平方近似的谱图网络[J]. 山东大学学报 (工学版), 2024, 54(5): 93-100.
[6] 常新功,苏敏惠,周志刚. 基于进化集成的图神经网络解释方法[J]. 山东大学学报 (工学版), 2024, 54(4): 1-12.
[7] 李璐,张志军,范钰敏,王星,袁卫华. 面向冷启动用户的元学习与图转移学习序列推荐[J]. 山东大学学报 (工学版), 2024, 54(2): 69-79.
[8] 赵涛,张宁,王小超,马川义,田源,张圣涛,杨梓梁. 基于图神经网络轨迹预测的合流区交通冲突预测方法[J]. 山东大学学报 (工学版), 2024, 54(2): 36-46.
[9] 陈雷,赵耀帅,林彦,郭晟楠,万怀宇,林友芳. 交通流量预测的时间异质性图注意力网络[J]. 山东大学学报 (工学版), 2023, 53(5): 29-36.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
版权所有 © 2018 《山东大学学报(工学版)》编辑部　
地址: 济南市山大南路27号（250100）　 电话: 0531-88366735　 E-mail: xbgxb@sdu.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发