计算机科学

计算机科学 ›› 2024, Vol. 51 ›› Issue (8): 304-312.doi: 10.11896/jsjkx.240100139

• 人工智能 • 上一篇    下一篇

基于标签传播增强的多通道图卷积网络

袁立宁1,2, 冯文刚1, 刘钊3   

  1. 1 中国人民公安大学国家安全学院 北京 100038
    2 广西警察学院公安大数据现代产业学院 南宁 530028
    3 中国人民公安大学研究生院 北京 100038
  • 收稿日期:2024-01-17 修回日期:2024-05-07 出版日期:2024-08-15 发布日期:2024-08-13
  • 通讯作者: 冯文刚(fengwengang@ppsuc.edu.cn)
  • 作者简介:(yuanlining@gxjcxy.edu.cn)
  • 基金资助:
    国家重点研发计划(2023YFC3321604);广西哲学社会科学研究课题(23FTQ005);北京市社会科学基金(22GLB225);广西壮族自治区公安厅专项课题(2023GAQN092).

Multi-channel Graph Convolutional Networks Enhanced by Label Propagation Algorithm

YUAN Lining1,2, FENG Wengang1, LIU Zhao3   

  1. 1 School of National Security,People’s Public Security University of China,Beijing 100038,China
    2 School of Public Security Big Data Modern Industry,Guangxi Police College,Nanning 530028,China
    3 Graduate School,People’s Public Security University of China,Beijing 100038,China
  • Received:2024-01-17 Revised:2024-05-07 Online:2024-08-15 Published:2024-08-13
  • About author:YUAN Lining,born in 1995,doctoral student,is a member of CCF(No.H5844M).His main research interests include machine learning and graph neural network.
    FENG Wengang,born in 1982,Ph.D,professor,Ph.D supervisor.His main research interests include pattern recognition and public security intelligence.
  • Supported by:
    National Key Research and Development Program of China(2023YFC3321604),Social Science Fund of Guangxi(23FTQ005),Social Science Fund of Beijing(22GLB225) and Special Fund of Guangxi Public Security Department(2023GAQN092).

PDF (PC)

318

摘要： 多数图卷积网络(GCN)模型通过设计高效的信息传递和保留方式提升节点分类任务的实验表现,忽略了节点标签信息在拓扑空间和属性空间的传播。针对上述问题,提出了一种基于标签传播算法(LPA)增强的多通道图卷积模型MGCN-LPA,同时增大同类节点在属性和拓扑空间的关系权重,改善节点间特征和标签信息的传播。首先,计算不同节点的属性相似度值,并采用k近邻算法生成属性关系图;然后,利用结合了GCN和LPA的图卷积层GCN-LPA提取属性图和属性关系图的潜在特征,生成拓扑节点表示和属性节点表示;最后,将拓扑和属性表示进行融合,并将生成的最终表示用于节点分类任务。在3个基准图数据集上进行实验,MGCN-LPA的实验表现能够匹配当前较为先进的基线模型,其在Cora和Citeseer数据集上的分类结果相比表现最优的基线模型提升了9.3%和12%。上述实验结果表明,MGCN-LPA能够增大同类节点间路径的权重,从而增强同类节点间的信息传递,提升节点分类任务的实验表现。此外,消融实验结果表明,与仅使用拓扑空间或者属性空间信息的变体相比,融合两类信息的MGCN-LPA能够充分提取和保留原始图中蕴含的潜在特征,提升模型的表征能力和泛化性。

关键词: 图卷积网络, 标签传播算法, 属性图, 属性关系图, 节点分类

Abstract: Most graph convolutional networks(GCN) improve the experimental performance of node classification tasks by designing efficient methods for information propagation and preservation,while ignoring the propagation of node label information in the topological and attribute spaces.Aiming at the above problems,the paper proposes a multi-channel graph convolution model MGCN-LPA enhanced by the label propagation algorithm(LPA).The model enhances the propagation of node features and label information by increasing the weights of relationship between nodes of the same class in the attribute space and topology space.Firstly,it calculates the similarity values of different node attributes and generates an attribute relation graph using the k-nearest neighbor algorithm.Then,it combines the GCN and LPA in the graph convolution layer GCN-LPA to extract potential features from the attribute graph and attribute relation graph,generating topological node representations and attribute node representations.Finally,the method combines the topological and attribute representations and utilizes the final representation for node classification tasks.On three benchmark graph datasets,the experimental performance of MGCN-LPA can match the current state-of-the-art baseline models.The classification results on the Cora and Citeseer datasets show improvements of 9.3% and 12% respectively compared to the best-performing baseline.The experimental results demonstrate that MGCN-LPA can increase the weights of paths between nodes of the same class and enhance the propagation of information among nodes of the same class,thereby enhancing the performance of node classification tasks.In addition,the ablation experiments demonstrate that the fusion of both topological space and attribute space information in MGCN-LPA enhances the model’s representational capacity and ge-neralization compared to variants using only one type of information.This fusion allows for the full extraction and preservation of latent features present in the original graph.

Key words: Graph convolutional network, Label propagation algorithm, Attribute graph, Attribute relation graph, Node classification

中图分类号: 

  • TP183
[1]MCLAREN C D,BRUNER M W.Citation network analysis[J].International Review of Sport and Exercise Psychology,2022,15(1):179-198.
[2]GHAREHCHOPOGH F S.An improved Harris Hawks optimi-zation algorithm with multi-strategy for community detection in social network[J].Journal of Bionic Engineering,2023,20(3):1175-1197.
[3]ZOU M,GAN Z,CAO R,et al.Similarity-navigated graph neural networks for node classification[J].Information Sciences,2023,633:41-69.
[4]NASIRI E,BERAHMAND K,LI Y.Robust graph regularization nonnegative matrix factorization for link prediction in attributed networks[J].Multimedia Tools and Applications,2023,82(3):3745-3768.
[5]XU M.Understanding graph embedding methods and their applications[J].SIAM Review,2021,63(4):825-853.
[6]BHATTI U A,TANG H,WU G,et al.Deep learning withgraph convolutional networks:An overview and latest applications in computational intelligence[J/OL].International Journal of Intelligent Systems,2023:1-28.https://dl.acm.org/doi/abs/10.1155/2023/8342104.
[7]LI C H,ZHU X F.Dual-channel graph random convolutionalnetworks for semi-supervised node classification[J].Journal of Chinese Computer Systems,2023,44(8):1656-1664.
[8]GUO M H,XU T X,LIU J J,et al.Attention mechanisms incomputer vision:A survey[J].Computational Visual Media,2022,8(3):331-368.
[9]VELICKOVIC P,CUCURULL G,CASANOVA A,et al.Graph attention networks[EB/OL].[2023-12-12].https://arxiv.org/abs/1710.10903v3.
[10]LI B,JING B,TONG H,et al.Graph communal contrastivelearning[C]//Proceedings of the 22nd International Conference on World Wide Web.New York:ACM,2022:1203-1213.
[11]XUE L,NONG L P,ZHANG W H,et al.An improved graph convolution network semi-supervised node classification[J].Computer Applications and Software,2021,38(10):153-158,163.
[12]RAGHAVAN U N,ALBERT R,KUMARA S.Near linear time algorithm to detect community structures in large-scale networks[J].Physical Review E,2007,76(3):036106.
[13]WANG H,LESKOVEC J.Unifying Graph Convolutional Neural Networks and Label Propagation[EB/OL].[2023-12-12].https://arxiv.linfen3.top/abs/2002.06755v1.
[14]CUNNINGHAM P,DELANY S J.k-Nearest neighbor classi-fiers-A Tutorial[J].ACM Computing Surveys(CSUR),2021,54(6):1-25.
[15]LIU J Q,TU W X,ZHU E.Survey on graph convolutional neural network[J].Computer Engineering & Science,2023,45(8):1472-1481.
[16]ZHANG L Y,SUN H H,SHI B B.Review of node classification methods based on graph convolutional neural networks[J/OL].Computer Science,2023:1-19.http://kns.cnki.net/kcms/detail/50.1075.TP.20230925.1655.162.html.
[17]WU F,SOUZA A,ZHANG T,et al.Simplifying graph convolutional networks[C]//Proceedings of the 36th International Conference on Machine Learning.Cambridge:PMLR,2019:6861-6871.
[18]TAILOR S A,OPOLKA F,LIO P,et al.Do we need anisotropic graph neural networks?[C]//Proceedings of the 2022 International Conference on Learning Representations.Open-Review.net,2022:1-19.
[19]BRODY S,ALON U,YAHAV E.How attentive are graph attention networks?[C]//Proceedings of the 2022 International Conference on Learning Representations.OpenReview.net,2022:1-26.
[20]GUO X C,ZHANG W Y,XIA Z X.Two-way data augmentation graph convolutional networks[J].Computer Engineering and Design,2023,44(8):2345-2351.
[21]RANI V,NABI S T,KUMAR M,et al.Self-supervised Lear-ning:A Succinct Review[J].Archives of Computational Me-thods in Engineering,2023,30(4):2761-2775.
[22]LI Y Q,WANG J,WANG F,et al.Semi-Supervised Node Classification Algorithm Based on Hierarchical Contrastive Learning[J].Pattern Recognition and Artificial Intelligence,2023,36(8):712-720.
[23]ZHONG L,YANG J,CHEN Z,et al.Contrastive graph convolutional networks with generative adjacency matrix[J].IEEE Transactions on Signal Processing,2023,71:772-785.
[24]HUANG J,DU L,CHEN X,et al.Robust mid-pass filteringgraph convolutional networks[C]//Proceedings of the 2023 International World Wide Web Conference.New York:ACM,2023:328-338.
[25]NIE S M,DU Y H,LU T L,et al.Survey on local community detection algorithms[J].Software Guide,2020,19(5):271-275.
[26]WANG X,ZHU M,BO D,et al.Am-gcn:Adaptive multi-channel graph convolutional networks[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Disco-very & Data Mining.New York:ACM,2020:1243-1253.
[27]KIPF T N,WELLING M.Semi-supervised classification withgraph convolutional networks[EB/OL].[2023-12-19].https://arxiv.org/abs/1609.02907v4.
[28]YUAN L N,LI X,WANG X D,et al.Graph embedding models:a survey[J].Journal of Frontiers of Computer Science and Technology,2022,16(1):59-87.
[29]HU Y,YOU H,WANG Z,et al.Graph-mlp:node classification without message passing in graph[EB/OL].[2023-12-19].https://arxiv.org/abs/2106.04051v1.
[30]ZHENG W P,GE H L,LIU M L,et al.Node classification algorithm fusing 2-connected motif-structure information[J/OL].Journal of Computer Applications,2023:1-10.http://kns.cnki.net/kcms/detail/51.1307.tp.20230816.1507.006.html.
[31]ZENG J X,WANG P H,DING Y D,et al.Graph neural network based node embedding enhancement model for node classification[J].Journal of Zhe Jiang University(Engineering Science),2023,57(2):219-225.
[32]CAI T T,MA R.Theoretical foundations of t-SNE for visualizing high-dimensional clustered data[J].The Journal of Machine Learning Research,2022,23(1):13581-13634.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市渝北区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发