基于元路径卷积的异构图神经网络算法

doi:10.11959/j.issn.1000-0801.2024070

电信科学 ›› 2024, Vol. 40 ›› Issue (3): 89-103.doi: 10.11959/j.issn.1000-0801.2024070

• 研究与开发 • 上一篇

基于元路径卷积的异构图神经网络算法

秦志龙¹^,², 邓琨¹^,³, 刘星妍¹

¹ 嘉兴大学信息科学与工程学院，浙江嘉兴 314001
² 浙江理工大学计算机科学与技术学院（人工智能学院），浙江杭州 310018
³ 嘉兴大学浙江省全省多模态感知与智能系统重点实验室，浙江嘉兴 314001

修回日期:2024-03-15 出版日期:2024-03-01 发布日期:2024-03-01
作者简介:秦志龙（1999- ），男，浙江理工大学计算机科学与技术学院（人工智能学院）硕士生，主要研究方向为异构图神经网络和深度学习等
邓琨（1980- ），男，博士，嘉兴大学浙江省全省多模态感知与智能系统重点实验室、嘉兴大学信息科学与工程学院副教授、硕士生导师，主要研究方向为网络结构分析、数据挖掘、异构网络分析等
刘星妍（1980- ），女，嘉兴大学信息科学与工程学院高级工程师，主要研究方向为数据挖掘、网络结构分析等
基金资助:
教育部人文社会科学研究专项任务项目(22JDSZ3023);教育部产学合作协同育人项目(220603372015422);教育部产学合作协同育人项目(220604029012441)

Meta-path convolution based heterogeneous graph neural network algorithm

Zhilong QIN¹^,², Kun DENG¹^,³, Xingyan LIU¹

¹ College of Information Science and Engineering, Jiaxing University, Jiaxing 314001, China
² School of Computer Science and Technology (School of Artificial Intelligence), Zhejiang Sci-Tech University, Hangzhou 310018, China
³ Provincial Key Laboratory of Multimodal Perceiving and Intelligent Systems, Jiaxing University, Jiaxing 314001, China

Revised:2024-03-15 Online:2024-03-01 Published:2024-03-01
Supported by:
The Humanity and Social Science Research Project of Ministry of Education of China(22JDSZ3023);The Ministry of Education’s Industry-University Collaboration Education Project(220603372015422);The Ministry of Education’s Industry-University Collaboration Education Project(220604029012441)

摘要/Abstract

摘要：

现有异构图嵌入方法在多层图卷积计算中，通常将每个节点表示为单个向量，使得高阶图卷积层无法区分不同关系和顺序的信息，导致信息在传递过程中丢失。为解决该问题，提出了基于元路径卷积的异构图神经网络算法。该方法首先利用特征转换自适应调整节点特征；其次，设计了元路径内卷积挖掘节点高阶间接关系，捕获目标节点在单元路径下与其他类型节点之间的交互关系；最后，通过自注意力机制探索语义之间的相互性，融合来自不同元路径的特征。在ACM、IMDB和DBLP数据集上进行广泛实验，并与当前主流算法进行对比分析。实验结果显示，节点分类任务中Macro-F1平均提高0.5%～3.5%，节点聚类任务中ARI值提高了1%～3%，证明该算法是有效、可行的。

关键词: 异构图, 图嵌入, 图神经网络, 元路径, 图卷积

Abstract:

In the multilayer graph convolution calculation, each node is usually represented as a single vector, which makes the high-order graph convolution layer unable to distinguish the information of different relationships and sequences, resulting in the loss of information in the transmission process.To solve this problem, a heterogeneous graph neural network algorithm based on meta-path convolution was proposed.Firstly, the feature transformation was used to adaptively adjust the node features.Secondly, the high-order indirect relationship between the nodes was mined by convolution within the meta-path to capture the interaction between the target node and other types of nodes under the element path.Finally, the reciprocity between semantics was explored through the self-attention mechanism, and the features from different meta-paths were fused.Extensive experiments were carried out on ACM, IMDB and DBLP datasets, and compared with the current mainstream algorithms.The experimental results show that the average increase of Macro-F1 in the node classification task is 0.5%～3.5%, and the ARI value in the node clustering task is increased by 1%～3%, which proves that the algorithm is effective and feasible.

Key words: heterogeneous graph, graph embedding, graph neural network, meta-path, graph convolution

中图分类号:

TP391

秦志龙, 邓琨, 刘星妍. 基于元路径卷积的异构图神经网络算法[J]. 电信科学, 2024, 40(3): 89-103.

Zhilong QIN, Kun DENG, Xingyan LIU. Meta-path convolution based heterogeneous graph neural network algorithm[J]. Telecommunications Science, 2024, 40(3): 89-103.

图/表 11

图1

图2

图3

表1

表2

在IMDB、DBLP 和 ACM 数据集上的节点分类结果"

	指标	训练集比例	metapath2vec	GCN	HAN	MAGNN	HGNN_AC	HGB	OSGNN	MCHGNN
IMDB	Macro-F1	20%	46.3%	45.0%	57.4%	57.9%	59.7%	60.7%	59.5%	$61 . 2 %$
		40%	47.7%	45.4%	57.5%	59.1%	60.2%	61.2%	60.4%	$63 . 8 %$
		60%	48.2%	46.8%	57.7%	59.6%	60.6%	62.6%	60.8%	$65 . 1 %$
		80%	48.9%	47.2%	57.8%	59.8%	60.8%	63.2%	61.0%	$66 . 6 %$
	Micro-F1	20%	48.2%	46.3%	57.2%	58.0%	59.8%	60.8%	59.5%	$61 . 2 %$
		40%	49.5%	48.1%	57.6%	59.3%	60.4%	61.4%	60.4%	$63 . 8 %$
		60%	50.2%	48.5%	57.7%	59.7%	60.8%	62.8%	60.8%	$65 . 0 %$
		80%	50.8%	48.6%	57.8%	60.0%	61.0%	63.2%	61.0%	$66 . 6 %$
ACM	Macro-F1	20%	70.0%	70.2%	90.1%	88.9%	89.2%	91.1%	$92 . 6 %$	92.0%
		40%	71.3%	70.7%	90.2%	89.7%	90.7%	92.4%	$93 . 3 %$	92.2%
		60%	71.8%	70.6%	91.4%	90.2%	91.5%	92.6%	$93 . 7 %$	93.0%
		80%	72.3%	68.4%	91.5%	90.3%	91.9%	92.9%	93.8%	$94 . 0 %$
	Micro-F1	20%	72.3%	73.2%	89.9%	88.7%	89.1%	90.2%	$92 . 6 %$	92.2%
		40%	73.5%	73.6%	90.3%	89.2%	90.7%	91.3%	$93 . 1 %$	92.3%
		60%	73.9%	74.2%	91.4%	90.3%	91.5%	91.5%	93.6%	$93 . 7 %$
		80%	74.2%	73.9%	91.7%	90.8%	91.9%	92.3%	93.7%	$93 . 8 %$
DBLP	Macro-F1	20%	91.6%	90.1%	91.6%	92.4%	94.2%	94.0%	94.0%	$94 . 8 %$
		40%	92.8%	90.2%	91.6%	93.5%	94.4%	94.1%	94.3%	$95 . 0 %$
		60%	93.0%	90.4%	92.1%	93.9%	94.4%	94.2%	94.5%	$94 . 8 %$
		80%	93.2%	90.7%	92.2%	93.9%	94.6%	94.3%	94.8%	$95 . 3 %$
	Micro-F1	20%	92.2%	90.4%	92.3%	92.9%	94.6%	94.2%	94.4%	$95 . 3 %$
		40%	93.2%	90.9%	92.5%	93.0%	94.7%	94.5%	94.7%	$95 . 4 %$
		60%	93.9%	91.0%	92.6%	93.4%	94.8%	94.5%	94.9%	$95 . 2 %$
		80%	94.0%	91.1%	92.8%	93.9%	95.0%	94.6%	95.2%	$95 . 6 %$

表2

表3

节点聚类结果"

算法	IMDB		ACM		DBLP
算法	NMI	ARI	NMI	ARI	NMI	ARI
MAGNN	13.3%	11.6%	65.5%	69.3%	79.2%	84.6%
HGNN_AC	13.6%	14.4%	64.9%	68.5%	79.6%	84.9%
OSGNN	15.9%	18.1%	72.4%	74.5%	81.6%	86.4%
MCHGNN	$16 . 8 %$	$19 . 4 %$	$74 . 5 %$	$77 . 1 %$	$83 . 1 %$	$87 . 4 %$

表3

图4

图5

图6

表4

图7

参考文献 29

[1]	WEI Y C , FU X C , SUN Q Y ,et al. Heterogeneous graph neural network for privacy-preserving recommendation[C]// Proceedings of the 2022 IEEE International Conference on Data Mining (ICDM). Piscataway:IEEE Press, 2022: 528-537.
[2]	杨帅, 王瑞琴, 马辉 . 基于多通道的边学习图卷积网络[J]. 电信科学, 2022,38(9): 95-104.
	YANG S , WANG R Q , MA H . Multi-channel based edgelearning graph convolutional network[J]. Telecommunications Science, 2022,38(9): 95-104.
[3]	WANG J C , ZHANG Y , WANG L X ,et al. Multitask hypergraph convolutional networks:a heterogeneous traffic prediction framework[J]. IEEE Transactions on Intelligent Transportation Systems, 2022,23(10): 18557-18567.
[4]	刘群, 谭洪胜, 张优敏 ,等. 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 Q , TAN H S , ZHANG Y M ,et al. Dynamic heterogeneous network representation method based on meta-path[J]. Acta Electronica Sinica, 2022,50(8): 1830-1839.
[5]	YU P Y , FU C F , YU Y W ,et al. Multiplex heterogeneous graph convolutional network[C]// Proceedings of the Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York:ACM, 2022: 2377-2387.
[6]	陶涛, 李珍, 王冀彬 ,等. 基于图神经网络的权益推荐技术方案研究[J]. 电信科学, 2023,39(8): 91-101.
	TAO T , LI Z , WANG J B ,et al. Research on the graphical convolution neural network based benefits recommendation system strategy[J]. Telecommunications Science, 2023,39(8): 91-101.
[7]	KIPF T N , WELLING M . Semi-supervised classification with graph convolutional networks[J]. arXiv preprint,arXiv:1609.02907, 2016.
[8]	VELIKOVI P , CUCURULL G , CASANOVA A ,et al. Graph attention networks[J]. arXiv preprint,arXiv:1710.10903, 2017.
[9]	WANG X , JI H Y , SHI C ,et al. Heterogeneous graph attention network[C]// Proceedings of the WWW '19:The World Wide Web Conference. New York:ACM, 2019: 2022-2032.
[10]	FU X Y , ZHANG J N , MENG Z Q ,et al. MAGNN:metapath aggregated graph neural network for heterogeneous graph embedding[C]// Proceedings of the Proceedings of The Web Conference 2020. New York:ACM, 2020: 2331-2341.
[11]	YANG X C , YAN M Y , PAN S R ,et al. Simple and efficient heterogeneous graph neural network[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2023,37(9): 10816-10824.
[12]	ZHAO T Y , YANG C , LI Y B ,et al. Space4HGNN:a novel,modularized and reproducible platform to evaluate heterogeneous graph neural network[C]// Proceedings of the Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York:ACM, 2022: 2776-2789.
[13]	任嘉睿, 张海燕, 朱梦涵 ,等. 基于元图卷积的异质网络嵌入学习算法[J]. 计算机研究与发展, 2022,59(8): 1683-1693.
	REN J R , ZHANG H Y , ZHU M H ,et al. Embedding learning algorithm for heterogeneous network based on meta-graph convolution[J]. Journal of Computer Research and Development, 2022,59(8): 1683-1693.
[14]	JIN D , HUO C Y , LIANG C D ,et al. Heterogeneous graph neural network via attribute completion[C]// Proceedings of the Proceedings of the Web Conference 2021. New York:ACM, 2021: 391-400.
[15]	XU W T , XIA Y C , LIU W Q ,et al. Shgnn:structure-aware heterogeneous graph neural network[J]. arXiv preprint,arXiv:2112.06244, 2021.
[16]	ZHAN L , JIA T . CoarSAS2hvec:heterogeneous information network embedding with balanced network sampling[J]. Entropy, 2022,24(2): 276.
[17]	YUN S , JEONG M , KIM R ,et al. Graph transformer networks[J]. Advances in neural information processing systems, 2019,32.
[18]	YANG Y M , GUAN Z Y , LI J X ,et al. Interpretable and efficient heterogeneous graph convolutional network[J]. IEEE Transactions on Knowledge and Data Engineering, 2023,35(2): 1637-1650.
[19]	CHEN K J , LU H , LIU Z ,et al. Heterogeneous graph convolutional network with local influence[J]. Knowledge-Based Systems, 2022(236): 107699.
[20]	LYU Q S , DING M , LIU Q ,et al. Are we really making much progress? :Revisiting,benchmarking and refining heterogeneous graph neural networks[C]// Proceedings of the Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery ＆ Data Mining. New York:ACM, 2021: 1150-1160.
[21]	DU C G , YAO K C , ZHU H S ,et al. Seq-HGNN:learning sequential node representation on heterogeneous graph[J]. arXiv preprint,arXiv:2305.10771, 2023.
[22]	王悦天, 傅司超, 彭勤牧 ,等. 半监督场景下多视角信息交互的图卷积神经网络[J]. 软件学报, 2023: 1-18.
	WANG Y T , FU S C , PENG Q M ,et al. Multi-view interaction graph convolutional network for semi-supervised classification[J]. Journal of Software, 2023: 1-18.
[23]	WANG X , BO D Y , SHI C ,et al. A survey on heterogeneous graph embedding:methods,techniques,applications and sources[J]. IEEE Transactions on Big Data, 2023,9(2): 415-436.
[24]	GUAN W L , JIAO F K , SONG X M ,et al. Personalized fashion compatibility modeling via metapath-guided heterogeneous graph learning[C]// Proceedings of the Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York:ACM, 2022: 482-491.
[25]	WU Z Y , LIANG Q Y , ZHAN Z H . Course recommendation based on enhancement of meta-path embedding in heterogeneous graph[J]. Applied Sciences, 2023,13(4): 2404.
[26]	CHEN H , HONG P F , HAN W ,et al. Dialogue relation extraction with document-level heterogeneous graph attention networks[J]. Cognitive Computation, 2023,15(2): 793-802.
[27]	DONG Y X , CHAWLA N V , SWAMI A . metapath2vec:scalable representation learning for heterogeneous networks[C]// Proceedings of the Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York:ACM, 2017: 135-144.
[28]	YAN Y Y , LI C , YU Y W ,et al. OSGNN:original graph and subgraph aggregated graph neural network[J]. Expert Systems with Applications, 2023(225): 120115.
[29]	VAN DER M L , HINTON G . Visualizing data using t-SNE[J]. Journal of Machine Learning Research, 2008(9): 2579-2605.

数据集	节点类型	边类型	元路径
IMDB	M：4 287	M-D：4 278	M-D-M
	D：2 081	M-A：12 828	M-A-M
	A：5 257
ACM	P：4 287	P-P：9 615	P-A-P
	A：7 167	P-A：13 407	P-S-P
	S：60	P-S：4 019
DBLP	A：4 057	A-P：19 645	A-P-A
	P：14 328	P-T：85 810	A-P-T-P-A
	T：7 723	P-V：14 328	A-P-V-P-A
	V：20

元路径	节点类型	重要性权值
MDM	M	0.11×0.61=0.067 1
	MD	0.37×0.61=0.225 7
	MDM	0.52×0.61=0.317 2
MAM	M	0.20×0.39=0.078 0
	MA	0.42×0.39=0.163 8
	MAM	0.38×0.39=0.148 2

基于元路径卷积的异构图神经网络算法

Meta-path convolution based heterogeneous graph neural network algorithm

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 29

相关文章 7

Metrics

推荐阅读 0

[1]	王瑞琴, 黄熠旻, 纪其顺, 万超艺, 周志峰. 注意力感知的边-节点交换图神经网络模型[J]. 电信科学, 2024, 40(1): 106-114.
[2]	张雄涛, 郑景玉, 申情, 孙丹枫, 蒋云良. 基于混合图卷积的多通道时空交通流预测模型[J]. 电信科学, 2023, 39(9): 97-110.
[3]	陶涛, 李珍, 王冀彬, 徐海勇, 江勇, 陈卓, 张润波, 胡清源. 基于图神经网络的权益推荐技术方案研究[J]. 电信科学, 2023, 39(8): 91-101.
[4]	杨帅, 王瑞琴, 马辉. 基于多通道的边学习图卷积网络[J]. 电信科学, 2022, 38(9): 95-104.
[5]	李奕江, 叶会标, 谢仁华, 楼佳丽, 庄丹娜, 李传煌. 基于图神经网络的网络性能智能预测[J]. 电信科学, 2022, 38(3): 143-157.
[6]	邬少清,董一鸿,王雄,曹燕,辛宇. 基于高阶相似性的属性网络表示学习[J]. 电信科学, 2020, 36(12): 20-32.
[7]	王锐,张志强,石川. 异质信息网络分析及其语义探索[J]. 电信科学, 2015, 31(7): 43-51.