吉林大学学报(工学版) ›› 2021, Vol. 51 ›› Issue (3): 1048-1054.doi: 10.13229/j.cnki.jdxbgxb20200034

• 计算机科学与技术 • 上一篇    

融合全局和局部特征的胶囊图神经网络

钱榕1,2(),张茹2,张克君1,2,金鑫1,葛诗靓2,江晟3,4()   

  1. 1.北京电子科技学院 研究生部,北京 100070
    2.西安电子科技大学 计算机科学与技术学院,西安 710071
    3.长光卫星技术有限公司,长春 130000
    4.中国科学院 长春光学精密机械与物理研究所,长春 130033
  • 收稿日期:2020-01-14 出版日期:2021-05-01 发布日期:2021-05-07
  • 通讯作者: 江晟 E-mail:rqian@besti.edu.cn;jiangsheng10@mails.jlu.edu.cn
  • 作者简介:钱榕(1970-),男,副教授,博士. 研究方向:复杂网络,数据挖掘,云计算安全. E-mail:rqian@besti.edu.cn
  • 基金资助:
    国家重点研发计划项目(2018YFB1004101)

Capsule graph neural network based on global and local features fusion

Rong QIAN1,2(),Ru ZHANG2,Ke-jun ZHANG1,2,Xin JIN1,Shi-liang GE2,Sheng JIANG3,4()   

  1. 1.College of Graduate,Beijing Electronic Science and Technology Institute,Beijing 100070,China
    2.College of Computer Science and Technology,Xidian University,Xi'an 710071,China
    3.Chang Guang Satellite Technology Co. ,Ltd. ,Changchun 130000,China
    4.Changchun Institute of Optics,Fine Mechanics and Physics,Chinese Academy of Sciences,Changchun 130033,China
  • Received:2020-01-14 Online:2021-05-01 Published:2021-05-07
  • Contact: Sheng JIANG E-mail:rqian@besti.edu.cn;jiangsheng10@mails.jlu.edu.cn

摘要:

针对胶囊图神经网络训练中得到的只有整体结构信息,并且随着层数的增加,节点的结构特征信息会丢失的问题,本文提出融合全局和局部特征的胶囊图神经网络。首先,改进了Node2vec,将节点的属性信息引入随机游走过程中,从而在生成网络表示时综合考虑了网络结构和节点的属性;然后,将改进的Node2vec引入胶囊图神经网络,设计了一个融合全局和局部特征的胶囊图神经网络。通过实验发现,本文模型在训练时的收敛速度更快,在图分类任务上的准确率有所提高。

关键词: 计算机应用技术, 网络表示学习, 复杂网络, 图神经网络

Abstract:

The overall structure information is obtained in the training of the capsule graph neural network, and as the layers increases, the structure feature information of the node will be lost. A capsule graph neural network that combines global and local features was proposed. First, the Node2vec is improved, and the attribute information of nodes is introduced into the random walk process, so that the network structure and the attributes of nodes are taken into account when the network representation is generated. Then, the improved Node2vec is introduced into the capsule graph neural network, and the capsule graph neural network is designed which fuses global and local characteristics. Experimental results show that the proposed capsule graph neural network has faster training convergence, and higher graph classification accuracy.

Key words: computer application technology, network representation learning, complex network, graph neural network

中图分类号: 

  • TP181

图1

融合全局和局部特征的胶囊图神经网络"

图2

Node2vec的随机游走"

图3

图卷积模块"

图4

注意力模块"

图5

重建损失"

表1

数据集详细信息"

数据集数据大小标签平均节点数平均边个数
COLLAB3500074.494914.99
IMDB-B2100019.77193.06
IMDB-M3150013131.87
D&D21178284.31715.65
ENZYMES660032.4663.14

表2

对比实验准确率 (%)"

算法COLLABIMDB-BIMDB-MD&DENZYMES
WL79.02±1.7773.40±4.6349.33±4.7579.78±0.3652.22±1.26
GK72.84±0.2865.87±0.9843.89±0.3878.45±0.2632.70±1.20
AWE73.93±1.9474.45±5.8351.54±3.6171.51±4.0235.77±5.93
DGCNN73.76±0.4970.03±0.8647.83±0.8579.37±0.9451.00±7.29
CapsGNN79.62±0.9173.10±4.8350.27±2.6575.38±4.1754.67±5.67
GLCapsGNN77.88±4.1975.62±5.3863.12±2.2674.88±3.4258.43±4.25

表3

对比实验的计算代价"

算法COLLABIMDB-BIMDB-MD&DENZYMES
时间/s内存/%时间/s内存/%时间/s内存/%时间/s内存/%时间/s内存/%
CapsGNN0.5463630.1608590.1574576.1603872.273871
GLCapsGNN0.5792610.1945590.1753586.6494892.345272

图6

两种模型在IMDB-M数据集上的准确率对比曲线"

1 Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
2 Bruna J, Zaremba W, Szlam A, et al. Spectral networks and locally connected networks on graphs[J/OL]. [2019-01-02].
3 Defferrard M, Bresson X, Vandergheynst P. Convolutional neural networks on graphs with fast localized spectral filtering[C]∥Advances in Neural Information Processing Systems, San Francisco,2016:3844-3852.
4 Kipf T N, Welling M. Semi-supervised classification with graph convolutional networks[J/OL]. [2020-01-04].
5 Hamilton W, Ying Z, Leskovec J. Inductive representation learning on large graphs[C]∥Proceedings of the 31st International Conference on Neural Information Processing, San Francisco, 2017:1025-1035.
6 Verma S, Zhang Z L. Graph capsule convolutional neural networks[J/OL]. [2020-01-02].
7 Mallea M D G, Meltzer P, Bentley P J. Capsule neural networks for graph classification using explicit tensorial graph representations[J/OL]. [2020-01-04].
8 Zhu Z, Peng G, Chen Y, et al. A convolutional neural network based on a capsule network with strong generalization for bearing fault diagnosis[J]. Neurocomputing,2019,323: 62-75.
9 Grover A, Leskovec J. Node2vec: scalable feature learning for networks[C]∥Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, New York, 2016: 855-864.
10 Perozzi B, Al-Rfou R, Skiena S. Deepwalk: online learning of social representations[C]∥Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, New York, 2014:701-710.
11 Ribeiro L F R, Saverese P H P, Figueiredo D R. Struc2vec: learning node representations from structural identity[C]∥Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, New York, 2017:385-394.
12 Tang J, Qu M, Wang M, et al. LINE: large-scale information network embedding[C]∥Proceedings of the 24th International Conference on World Wide Web, Piscataway, 2015:1067-1077.
13 Henaff M, Bruna J, LeCun Y. Deep convolutional networks on graph-structured data[J/OL]. [2019-01-06].
14 Sabour S, Frosst N, Hinton G E. Dynamic routing between capsules[J/OL]. [2020-01-08].
15 Shervashidze N, Schweitzer P, van Leeuwen E J, et al. Weisfeiler-lehman graph kernels[J]. Journal of Machine Learning Research,2011,1(3):2539-2561.
16 Rezaee B. A cluster validity index for fuzzy clustering[J]. Fuzzy Sets and Systems, 2010, 161(23):3014-3025.
17 Wang H, Fan W, Yu P S, et al. Mining concept-drifting data streams using ensemble classifiers[C]∥Proceedings of the 9th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, New York, 2003:226-235.
18 Wang Y, Sun Y, Liu Z, et al. Dynamic graph CNN for learning on point clouds[J]. ACM Transactions on Graphics,2019,38(5):1-12.
[1] 周炳海,吴琼. 基于多目标的机器人装配线平衡算法[J]. 吉林大学学报(工学版), 2021, 51(2): 720-727.
[2] 许骞艺,秦贵和,孙铭会,孟诚训. 基于改进的ResNeSt驾驶员头部状态分类算法[J]. 吉林大学学报(工学版), 2021, 51(2): 704-711.
[3] 魏晓辉,孙冰怡,崔佳旭. 基于图神经网络的兴趣活动推荐算法[J]. 吉林大学学报(工学版), 2021, 51(1): 278-284.
[4] 宋元,周丹媛,石文昌. 增强OpenStack Swift云存储系统安全功能的方法[J]. 吉林大学学报(工学版), 2021, 51(1): 314-322.
[5] 车翔玖,董有政. 基于多尺度信息融合的图像识别改进算法[J]. 吉林大学学报(工学版), 2020, 50(5): 1747-1754.
[6] 马健, 樊建平, 刘峰, 李红辉. 面向对象软件系统演化模型[J]. 吉林大学学报(工学版), 2018, 48(2): 545-550.
[7] 时小虎, 冯国香, 李牧, 李瑛, 吴春国. 基于密度峰值的重叠社区发现算法[J]. 吉林大学学报(工学版), 2017, 47(1): 242-248.
[8] 黄岚, 李玉, 王贵参, 王岩. 基于点距离和密度峰值聚类的社区发现方法[J]. 吉林大学学报(工学版), 2016, 46(6): 2042-2051.
[9] 郭玉泉, 李雄飞. 复杂网络社区的分形聚类检测方法[J]. 吉林大学学报(工学版), 2016, 46(5): 1633-1638.
[10] 胡冠宇, 乔佩利. 基于云群的高维差分进化算法及其在网络安全态势预测上的应用[J]. 吉林大学学报(工学版), 2016, 46(2): 568-577.
[11] 佟金, 王亚辉, 樊雪梅, 张书军, 陈东辉. 生鲜农产品冷链物流状态监控信息系统[J]. 吉林大学学报(工学版), 2013, 43(06): 1707-1711.
[12] 叶育鑫, 赵建民, 莫毓昌, 欧阳丹彤, 刘华文. 基于典型相关分析的复杂网络模块挖掘算法[J]. 吉林大学学报(工学版), 2013, 43(02): 424-428.
[13] 刘大有, 杨建宁, 杨博, 赵学华, 金弟. 基于环路紧密度的复杂网络社区挖掘方法[J]. 吉林大学学报(工学版), 2013, 43(01): 98-105.
[14] 金鑫, 谢斌, 朱建明. 基于复杂网络分析的微博网络舆情传播[J]. 吉林大学学报(工学版), 2012, 42(增刊1): 271-275.
[15] 刘大有1,2, 张冬威1,2,李妮娅1,2,刘杰1,2,金弟1,2. 基于网络聚类选择的神经网络集成方法及应用[J]. 吉林大学学报(工学版), 2011, 41(4): 1034-1040.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!