基于元路径的动态异质网络表示学习

doi:10.12263/DZXB.20211288

电子学报 ›› 2022, Vol. 50 ›› Issue (8): 1830-1839.DOI: 10.12263/DZXB.20211288

基于元路径的动态异质网络表示学习

刘群, 谭洪胜, 张优敏, 王国胤

重庆邮电大学计算机科学与技术学院，重庆 400065

收稿日期:2021-09-17 修回日期:2022-04-21 出版日期:2022-08-25
- 作者简介:
- 刘　群　女，1969年出生于江西省南昌市，博士，重庆邮电大学教授，博士生导师.主要研究方向为复杂网络、可解释人工智能和数据挖掘.E-mail： liuqun@cqupt.edu.cn
  谭洪胜　男，1996年出生于重庆市.硕士生，主要研究方向为网络表示学习.E-mail： redtan2000@163.com
  张优敏　女，1987年出生于陕西省兴平市.博士生，主要研究方向为图神经网络的可解释.E-mail： ymzhang0103@hotmail.com
  王国胤　男，1970年出生于重庆市，博士，重庆邮电大学教授，博士生导师.教育部“长江学者”特聘教授（2015‒2019）、中组部“万人计划”科技创新领军人才（2014）、人社部“新世纪百千万人才工程”国家级人选、国务院特殊津贴专家、中科院“百人计划”专家、教育部“新世纪优秀人才”.主要研究方向为粗糙集、粒计算、数据挖掘、认知计算、大数据、人工智能等.E-mail： wanggy@cqupt.edu.cn
- 基金资助:
- 国家自然科学基金重点项目 (61936001); 重庆市教委重点合作项目 (HZ2021008); 重庆市自然科学基金 (cstc2021ycjh-bgzxm0013)

Dynamic Heterogeneous Network Representation Method Based on Meta-Path

LIU Qun, TAN Hong-sheng, ZHANG You-min, WANG Guo-yin

College of Computer Science and Technology，Chongqing University of Posts and Telecommunications，Chongqing 400065，China

Received:2021-09-17 Revised:2022-04-21 Online:2022-08-25 Published:2022-09-08

摘要/Abstract

摘要：

对网络表示学习的研究已经取得了很多成果，但是大部分网络表示学习模型忽略了网络动态性和异质性，无法区分网络中耦合的时间和空间（结构）特征，也不能捕获网络的丰富语义信息.本文提出了基于元路径的动态异质网络表示学习方法.首先将节点的邻域结构按照时间划分出不同的子空间结构，并为每个节点采样出所有时间加权元路径的序列.其次通过门控循环单元将节点的全部时间加权元路径序列上的邻域信息进行集成，最后利用带注意力机制的双向门控循环单元对融合后的节点序列进行时空上下文信息学习，获得每个节点的最终表示向量.通过在真实数据集上的实验表明，在节点分类、聚类和可视化的下游任务测试中，本文提出的算法较基线方法在性能上均有较大提升.节点分类任务中的Micro-F1平均提高了1.09%~3.72%，节点聚类任务中的ARI值提高了3.23%～14.49%.

关键词: 网络表示学习, 动态异质网络, 元路径, 注意力机制, 门控循环单元

Abstract:

The researches of network representation learning have made many achievements. Since most of the researches ignore the dynamics and heterogeneity of the networks, coupled temporal and spatial structure features can not be distinguished, and rich semantic information of the network cannot be captured well. In this paper, meta-path based dynamic heterogeneous network representation learning method is proposed. Firstly, the neighborhood structures of nodes are divided into different sub-spaces according to their time, then the sequences of all time-weighted meta-paths for each node are sampled. Secondly, the neighborhood information on all time-weighted meta-paths of each node is integrated by a gated recurrent unit network(GRU). Furthermore, a bi-directional gated recurrent unit network(Bi-GRU) with an attention mechanism is used to learn the spatio-temporal contextual information from the merged sequences, and the final node representation will be received. Experiments on real data sets show that our algorithm has greatly improved performance on the downstream network tasks, such as node classification, clustering and visualization. Compared with state-of-the-art baseline methods,the Micro-F1 value has been raised by 1.09%~3.72% averagely on classification tasks, and the ARI value has been increased by 3.23%~14.49% on clustering tasks.

Key words: network representation learning, dynamic heterogeneous network, metapath, attention mechanism, gated recurrent unit

中图分类号:

TP391

刘群, 谭洪胜, 张优敏, 王国胤. 基于元路径的动态异质网络表示学习[J]. 电子学报, 2022, 50(8): 1830-1839.

LIU Qun, TAN Hong-sheng, ZHANG You-min, WANG Guo-yin. Dynamic Heterogeneous Network Representation Method Based on Meta-Path[J]. Acta Electronica Sinica, 2022, 50(8): 1830-1839.

图/表 11

参考文献 24

1	JI M, HAN J W, DANIEVSKY M. Ranking-based classification of heterogeneous information networks[C]//Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Diego: ACM, 2011: 1298-1306.
2	OPSAHL T, PANZARAS P. Clustering in weighted networks[J]. Social Networks, 2009, 31(2): 155-163.
3	DONG Y, CHAWLA N V, SWAMI A. Metapath2vec: Scalable representation learning for heterogeneous networks[C]//Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Halifax: ACM, 2017: 135-144.
4	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: ACM, 2014: 701-710.
5	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. San Francisco: ACM, 2016: 855-864.
6	MIKOLOV T, SUTSKEVER I, CHEN K, et al. Distributed representations of words and phrases and their compositionality[J]. Advances in Neural Information Processing Systems, 2013, 26: 3111-3119.
7	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. Florence:ACM, 2015: 1067-1077.
8	Wang D, Cui P, Zhu W. Structural deep network embedding[C]// Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: ACM, 2016: 1225-1234.
9	GORI M, MONFARDINI G, SCARSELLI F. A new model for learning in graph domains[C]//Proceedings of 2005 IEEE International Joint Conference on Neural Networks. Montreal: IEEE, 2005: 729-734.
10	WANG X, JI H, SHI C, et al. Heterogeneous graph attention network[C]//The World Wide Web Conference. San Francisco: .
11	QIAO Z, WANG P, FU Y, et al. Tree structure-aware graph representation learning via integrated hierarchical aggregation and relational metric learning[C]//20th IEEE International Conference on Data Mining. Sorrento: IEEE, 2020: 432-441.
12	ZHANG C, SONG D, HUANG C, et al. Heterogeneous graph neural network[C]//Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. Anchorage: ACM, 2019: 793-803.
13	KIPF T N, WELLING M. Semi-supervised classification with graph convolutional networks[EB/OL]. (2017-02-22)[2021-03-01]. .
14	ZHOU L, YANG Y, REN X, et al. Dynamic network embedding by modeling triadic closure process[C]//Proceedings of the AAAI Conference on Artificial Intelligence. New Orleans: AAAI, 2018, 32(1): 571-578.
15	尹赢, 张建朋, 吉立新, 李治成. 基于霍克斯点过程的动态网络表示学习方法[J]. 电子学报, 2020, 48(11): 2154-2161.
	YIN Y, ZHANG J P, JI L X, Li Z C. Dynamic network representation learning based onhawkes point process[J]. Acta Electronica Sinica, 2020, 48(11): 2154-2161. (in Chinese)
16	NGUYEN G H, LEE J B, ROSSI R A, et al. Continuous-time dynamic network embeddings[C]//Companion Proceedings of the The Web Conference 2018. Lyon: .
17	LU Y, WANG X, SHI C, et al. Temporal network embedding with micro-and macro-dynamics[C]//Proceedings of the 28th ACM International Conference on Information and Knowledge Management. Beijing:ACM, 2019: 469-478.
18	YIN Y, JI L X, ZHANG J P, et al. Dhne: Network representation learning method for dynamic heterogeneous networks[J]. IEEE Access, 2019: 134782-134792.
19	WANG X, LU Y, SHI C, et al. Dynamic heterogeneous information network embedding with meta-path based proximity[J]. IEEE Transactions on Knowledge and Data Engineering, 2020, 34(3): 1117-1132.
20	XUE H, YANG L, JIANG W, et al. Modeling dynamic heterogeneous network for link prediction using hierarchical attention with temporal RNN[C]//Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Ghent: Springer, 2020: 282-298.
21	SUN Y, HAN J, YAN X, et al. Pathsim: Meta path-based top-k similarity search in heterogeneous information networks[J]. Proceedings of the VLDB Endowment, 2011, 4(11): 992-1003.
22	SHI C, ZHANG Z, LUO P, et al. Semantic path based personalized recommendation on weighted heterogeneous information networks[C]//Proceedings of the 24th ACM International on Conference on Information and Knowledge Management. Shanghai: ACM, 2015: 453-462.
23	HU Z, DONG Y, WANG K, et al. Heterogeneous graph transformer[C]//Proceedings of the Web Conference 2020. Taipei: ACM, 2020: 2704-2710.
24	LE Q, MIKOLOV T. Distributed representations of sentences and documents[C]//International Conference on Machine Learning. Beijing: ACM, 2014: 1188-1196.

数据集	节点类型	节点数	时间	标签	元路径
AMiner	Paper(P)	18 181	16	5	APA APCPA
	Conference(C)	22
	Author(A)	22 942
DBLP-4	Paper(P)	14 328	4	4	APA APCPA APTPA
	Conference(C)	20
	Author(A)	4 057
	Term(T)	8 898
DBLP-10	Paper(P)	8 970	6	10	PAP PCP
	Conference(C)	51
	Author(A)	15 019

数据集	节点类型	节点数	时间	标签	元路径
AMiner	Paper(P)	18 181	16	5	APA APCPA
	Conference(C)	22
	Author(A)	22 942
DBLP-4	Paper(P)	14 328	4	4	APA APCPA APTPA
	Conference(C)	20
	Author(A)	4 057
	Term(T)	8 898
DBLP-10	Paper(P)	8 970	6	10	PAP PCP
	Conference(C)	51
	Author(A)	15 019

数据集	分类指标	训练比例	DeW	GCN	M2v	HAN	M²DNE	DHNE	DyHATR	DHNR_stu	DHNR
AMiner	Macro-F1	20%	0.969 8	0.972 7	0.974 1	0.960 5	0.965 2	0.845 3	0.971 0	0.972 7	0.9762
		50%	0.971 1	0.974 4	0.974 3	0.962 1	0.965 7	0.844 9	0.972 1	0.974 1	0.9768
		80%	0.970 1	0.974 7	0.974 8	0.964 4	0.967 3	0.846 1	0.972 8	0.973 4	0.9771
	Micro-F1	20%	0.972 1	0.974 9	0.976 5	0.964 2	0.967 3	0.853 9	0.972 7	0.975 2	0.9775
		50%	0.973 4	0.975 8	0.976 7	0.966 3	0.967 9	0.855 6	0.973 5	0.975 8	0.9788
		80%	0.972 5	0.976 6	0.977 4	0.968 1	0.969 1	0.855 5	0.974 9	0.975 6	0.9793
DBLP-4	Macro-F1	20%	0.876 9	0.909 3	0.923 2	0.925 1	0.9401	0.915 7	0.922 7	0.934 1	0.939 1
		50%	0.907 4	0.918 1	0.922 9	0.926 2	0.941 3	0.891 4	0.933 4	0.936 7	0.9431
		80%	0.920 2	0.923 7	0.923 3	0.925 7	0.941 0	0.907 8	0.936 9	0.938 8	0.9439
	Micro-F1	20%	0.885 8	0.915 1	0.928 5	0.930 1	0.9422	0.924 5	0.926 1	0.937 7	0.941 2
		50%	0.915 4	0.914 4	0.927 9	0.931 4	0.943 4	0.910 3	0.935 2	0.941 5	0.9486
		80%	0.926 2	0.919 7	0.928 0	0.941 0	0.944 6	0.926 2	0.944 3	0.943 6	0.9488
DBLP-10	Macro-F1	20%	0.822 2	0.841 9	0.854 4	0.841 5	0.863 4	0.788 1	0.871 5	0.881 2	0.8972
		50%	0.856 4	0.847 6	0.871 3	0.882 7	0.886 2	0.792 4	0.883 5	0.895 2	0.8984
		80%	0.866 8	0.863 1	0.877 8	0.889 2	0.890 1	0.809 2	0.889 5	0.896 1	0.8993
	Micro-F1	20%	0.838 7	0.874 3	0.862 1	0.885 1	0.881 8	0.777 1	0.872 4	0.9011	0.897 4
		50%	0.852 1	0.892 1	0.871 9	0.899 4	0.887 1	0.784 1	0.891 0	0.903 8	0.9073
		80%	0.856 6	0.896 4	0.880 5	0.901 2	0.892 3	0.821 8	0.899 2	0.909 4	0.9101

数据集	分类指标	训练比例	DeW	GCN	M2v	HAN	M²DNE	DHNE	DyHATR	DHNR_stu	DHNR
AMiner	Macro-F1	20%	0.969 8	0.972 7	0.974 1	0.960 5	0.965 2	0.845 3	0.971 0	0.972 7	0.9762
		50%	0.971 1	0.974 4	0.974 3	0.962 1	0.965 7	0.844 9	0.972 1	0.974 1	0.9768
		80%	0.970 1	0.974 7	0.974 8	0.964 4	0.967 3	0.846 1	0.972 8	0.973 4	0.9771
	Micro-F1	20%	0.972 1	0.974 9	0.976 5	0.964 2	0.967 3	0.853 9	0.972 7	0.975 2	0.9775
		50%	0.973 4	0.975 8	0.976 7	0.966 3	0.967 9	0.855 6	0.973 5	0.975 8	0.9788
		80%	0.972 5	0.976 6	0.977 4	0.968 1	0.969 1	0.855 5	0.974 9	0.975 6	0.9793
DBLP-4	Macro-F1	20%	0.876 9	0.909 3	0.923 2	0.925 1	0.9401	0.915 7	0.922 7	0.934 1	0.939 1
		50%	0.907 4	0.918 1	0.922 9	0.926 2	0.941 3	0.891 4	0.933 4	0.936 7	0.9431
		80%	0.920 2	0.923 7	0.923 3	0.925 7	0.941 0	0.907 8	0.936 9	0.938 8	0.9439
	Micro-F1	20%	0.885 8	0.915 1	0.928 5	0.930 1	0.9422	0.924 5	0.926 1	0.937 7	0.941 2
		50%	0.915 4	0.914 4	0.927 9	0.931 4	0.943 4	0.910 3	0.935 2	0.941 5	0.9486
		80%	0.926 2	0.919 7	0.928 0	0.941 0	0.944 6	0.926 2	0.944 3	0.943 6	0.9488
DBLP-10	Macro-F1	20%	0.822 2	0.841 9	0.854 4	0.841 5	0.863 4	0.788 1	0.871 5	0.881 2	0.8972
		50%	0.856 4	0.847 6	0.871 3	0.882 7	0.886 2	0.792 4	0.883 5	0.895 2	0.8984
		80%	0.866 8	0.863 1	0.877 8	0.889 2	0.890 1	0.809 2	0.889 5	0.896 1	0.8993
	Micro-F1	20%	0.838 7	0.874 3	0.862 1	0.885 1	0.881 8	0.777 1	0.872 4	0.9011	0.897 4
		50%	0.852 1	0.892 1	0.871 9	0.899 4	0.887 1	0.784 1	0.891 0	0.903 8	0.9073
		80%	0.856 6	0.896 4	0.880 5	0.901 2	0.892 3	0.821 8	0.899 2	0.909 4	0.9101

对比算法	AMiner		DBLP-4		DBLP-10
对比算法	NMI	ARI	NMI	ARI	NMI	ARI
DeW	0.782	0.691	0.761	0.811	0.404	0.212
GCN	0.879	0.907	0.724	0.770	0.396	0.483
M2v	0.727	0.591	0.780	0.830	0.294	0.171
HAN	0.787	0.810	0.781	0.835	0.430	0.540
M²DNE	0.735	0.657	0.756	0.786	0.306	0.195
DHNE	0.174	0.456	0.238	0.093	0.152	0.061
DyHATR	0.823	0.870	0.768	0.822	0.376	0.419
DHNR_stu	0.911	0.947	0.784	0.843	0.382	0.551
DHNR	0.916	0.943	0.810	0.862	0.413	0.553

基于元路径的动态异质网络表示学习

Dynamic Heterogeneous Network Representation Method Based on Meta-Path

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 24

相关文章 15

编辑推荐

Metrics

[1]	张志昌, 于沛霖, 庞雅丽, 朱林, 曾扬扬. SMGN：用于对话状态跟踪的状态记忆图网络[J]. 电子学报, 2022, 50(8): 1851-1858.
[2]	赵俊男, 佘青山, 孟明, 陈云. 基于多流空间注意力图卷积SRU网络的骨架动作识别[J]. 电子学报, 2022, 50(7): 1579-1585.
[3]	王延达, 陈炜通, 皮德常, 岳琳. 一种自适应记忆神经网络多跳读取与覆盖度机制结合的药物推荐模型[J]. 电子学报, 2022, 50(4): 943-953.
[4]	张凌明, 赵悦, 李鹏程, 刘洋, 高陈强. 基于局部注意力机制的三维牙齿模型分割网络[J]. 电子学报, 2022, 50(3): 681-690.
[5]	伍邦谷, 张苏林, 石红, 朱鹏飞, 王旗龙, 胡清华. 基于多分支结构的不确定性局部通道注意力机制[J]. 电子学报, 2022, 50(2): 374-382.
[6]	周登文, 李文斌, 李金新, 黄志勇. 一种轻量级的多尺度通道注意图像超分辨率重建网络[J]. 电子学报, 2022, 50(10): 2336-2346.
[7]	王波, 黄冕, 刘利军, 黄青松, 单文琦. 基于多层聚焦Inception-V3卷积网络的细粒度图像分类[J]. 电子学报, 2022, 50(1): 72-78.
[8]	赵琰, 赵凌君, 匡纲要. 基于注意力机制特征融合网络的SAR图像飞机目标快速检测[J]. 电子学报, 2021, 49(9): 1665-1674.
[9]	周东明, 张灿龙, 李志欣, 王智文. 基于多层级视觉融合的图像描述模型[J]. 电子学报, 2021, 49(7): 1286-1290.
[10]	孙全明, 曲志坚, 任崇广. 基于粒子群优化和LightGBM的情景感知多式联运推荐[J]. 电子学报, 2021, 49(5): 894-903.
[11]	李志欣, 孙亚茹, 唐素勤, 张灿龙, 马慧芳. 双路注意力引导图卷积网络的关系抽取[J]. 电子学报, 2021, 49(2): 315-323.
[12]	江泽涛, 钱艺, 伍旭, 张少钦. 一种基于ARD‑GAN的低照度图像增强方法[J]. 电子学报, 2021, 49(11): 2160-2165.
[13]	石义乐, 杨文忠, 杜慧祥, 王丽花, 王婷, 理珊珊. 基于深度学习的图像描述综述[J]. 电子学报, 2021, 49(10): 2048-2060.
[14]	唐海桃, 薛嘉宾, 韩纪庆. 一种多尺度前向注意力模型的语音识别方法[J]. 电子学报, 2020, 48(7): 1255-1260.
[15]	袁文浩, 胡少东, 时云龙, 李钊, 梁春燕. 一种用于语音增强的卷积门控循环网络[J]. 电子学报, 2020, 48(7): 1276-1283.