Subgraph-aware contrastive learning with data augmentation

doi:10.11772/j.issn.1001-9081.2025010110

Abstract

Abstract:

Graph Neural Network （GNN） is an effective graph representation learning method for processing graph structure data. However， the performance of GNN in practical applications is limited by the problem of missing information. On the one hand， the graph structure is usually sparse， making it difficult for the model to learn node features adequately. On the other hand， model training is limited because supervised learning relies on sparse label data， making it difficult to obtain robust node representations. To address these problems， a Subgraph-aware Contrastive Learning with Data Augmentation （SCLDA） model was proposed. Firstly， the relationship scores among nodes were obtained by learning the original graph through link prediction， and the edges with the highest scores were added to the original graph to generate the enhanced graph. Secondly， local subgraphs of the original and enhanced graphs were sampled by using target nodes respectively， and the target nodes of subgraphs were input to the shared GNN encoder， so as to generate the target node embeddings at subgraph level. Finally， the mutual information between similar instances was maximized on the basis of contrastive learning of the target nodes from the two perspective subgraphs. Experimental results of node classification on six public datasets Cora， Citeseer， Pubmed， Cora_ML， DBLP， and Photo show that SCLDA model improves the accuracy over the traditional GCN model by about 4.4%， 6.3%， 4.5%， 7.0%， 13.2% and 9.3%， respectively.

Key words: graph representation learning, Graph Neural Network (GNN), data augmentation, self-supervised learning, Graph Contrastive Learning (GCL), node classification

摘要：

图神经网络（GNN）是处理图结构数据的有效图表示学习方法。然而，在实际应用中， GNN的性能受限于信息缺失问题：一方面，图结构通常较为稀疏，导致模型难以充分学习节点特征；另一方面，监督学习依赖的标签数据通常稀缺，使模型训练受限，进而难以获得鲁棒的节点表示。针对以上问题，提出一种基于数据增强的子图感知对比学习（SCLDA）模型。首先，使用链路预测学习原始图得出节点之间的关系得分，并将得分最高的边添加到原始图中以生成增强图；其次，对原始图和增强图分别利用目标节点进行局部子图采样，将子图的目标节点输入共享GNN编码器，生成子图级别的目标节点嵌入；最后，基于2个视角子图的目标节点的对比学习最大化相似实例之间的互信息。在Cora、Citeseer、Pubmed、Cora_ML、DBLP和Photo 6个公共数据集上进行节点分类实验的结果表明， SCLDA模型比传统GCN模型的准确率分别提升了约4.4%、6.3%、4.5%、7.0%、13.2%和9.3%。

关键词: 图表示学习, 图神经网络, 数据增强, 自监督学习, 图对比学习, 节点分类

CLC Number:

TP18

Wen LI, Kairong LI, Kai YANG. Subgraph-aware contrastive learning with data augmentation[J]. Journal of Computer Applications, 2026, 46(1): 1-9.

李玟, 李开荣, 杨凯. 基于数据增强的子图感知对比学习[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 1-9.

Figures/Tables 9

References 55

[1]	MIN S， GAO Z， PENG J， et al. STGSN — a spatial-temporal graph neural network framework for time-evolving social networks ［J］. Knowledge-Based Systems， 2021， 214： No.106746.
[2]	DONG H， WANG P， XIAO M， et al. Temporal inductive path neural network for temporal knowledge graph reasoning ［J］. Artificial Intelligence， 2024， 329： No.104085.
[3]	ZHANG C， SONG D， HUANG C， et al. Heterogeneous graph neural network ［C］// Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York： ACM， 2019： 793-803.
[4]	WU Z， PAN S， CHEN F， et al. A comprehensive survey on graph neural networks ［J］. IEEE Transactions on Neural Networks and Learning Systems， 2021， 32（1）： 4-24.
[5]	ASIF N A， SARKER Y， CHAKRABORTTY R K， et al. Graph neural network： a comprehensive review on non-Euclidean space ［J］. IEEE Access， 2021， 9： 60588-60606.
[6]	FAN W， MA Y， LI Q， et al. Graph neural networks for social recommendation ［C］// Proceedings of the 2019 World Wide Web Conference. New York： ACM， 2019： 417-426.
[7]	刘渊，赵紫娟，杨凯.基于节点相似性的图注意力网络表示学习模型［J］.计算机应用研究， 2023， 40（3）： 822-827.
	LIU Y， ZHAO Z J， YANG K. Graph attention network representation learning with node similarity ［J］. Application Research of Computers， 2023， 40（3）： 822-827.
[8]	KIPF T N， WELLING M. Semi-supervised classification with graph convolutional networks ［EB/OL］. ［2024-12-20］. .
[9]	VELIČKOVIĆ P， CUCURULL G， CASANOVA A， et al. Graph attention networks ［EB/OL］. ［2024-12-20］. .
[10]	WANG R， MOU S， WANG X， et al. Graph structure estimation neural networks ［C］// Proceedings of the Web Conference 2021. New York： ACM， 2021： 342-353.
[11]	ZHAO T， LIU Y， NEVES L， et al. Data augmentation for graph neural networks ［C］// Proceedings of the 35th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2021： 11015-11023.
[12]	LI B， HOU Y， CHE W. Data augmentation approaches in natural language processing： a survey ［J］. AI Open， 2022， 3： 71-90.
[13]	SHORTEN C， KHOSHGOFTAAR T M. A survey on image data augmentation for deep learning ［J］. Journal of Big Data， 2019， 6： No.60.
[14]	YOU Y， CHEN T， SHEN Y， et al. Graph contrastive learning automated ［C］// Proceedings of the 38th International Conference on Machine Learning. New York： JMLR.org， 2021： 12121-12132.
[15]	YOU Y， CHEN T， SUI Y， et al. Graph contrastive learning with augmentations ［C］// Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2020： 5812-5823.
[16]	ZHU Y， XU Y， YU F， et al. Graph contrastive learning with adaptive augmentation ［C］// Proceedings of the Web Conference 2021. New York： ACM， 2021： 2069-2080.
[17]	HUANG J， CAO Q， XIE R， et al. Adversarial learning data augmentation for graph contrastive learning in recommendation ［C］// Proceedings of the 2023 International Conference on Database Systems for Advanced Applications， LNCS 13944. Cham： Springer， 2023： 373-388.
[18]	GILMER J， SCHOENHOLZ S S， RILEY P F， et al. Neural message passing for quantum chemistry ［C］// Proceedings of the 34th International Conference on Machine Learning. New York： JMLR.org， 2017： 1263-1272.
[19]	DUAN H， XIE C， LI B， et al. Self-supervised contrastive graph representation with node and graph augmentation ［J］. Neural Networks， 2023， 167： 223-232.
[20]	QIU J， CHEN Q， DONG Y， et al. GCC： graph contrastive coding for graph neural network pre-training ［C］// Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York： ACM， 2020： 1150-1160.
[21]	JIAO Y， XIONG Y， ZHANG J， et al. Sub-graph contrast for scalable self-supervised graph representation learning ［C］// Proceedings of the 2020 IEEE International Conference on Data Mining. Piscataway： IEEE， 2020： 222-231.
[22]	ZHENG C， ZONG B， CHENG W， et al. Robust graph representation learning via neural sparsification ［C］// Proceedings of the 37th International Conference on Machine Learning. New York： JMLR.org， 2020： 11458-11468.
[23]	DUAN J， WANG S， ZHANG P， et al. Graph anomaly detection via multi-scale contrastive learning networks with augmented view ［C］// Proceedings of the 37th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2023： 7459-7467.
[24]	TAO C， WANG H， ZHU X， et al. Exploring the equivalence of Siamese self-supervised learning via a unified gradient framework ［C］// Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2022： 14411-14420.
[25]	YANG K， LIU Y， ZHAO Z， et al. Local structure-aware graph contrastive representation learning ［J］. Neural Networks， 2024， 172： No.106083.
[26]	XIAO S， WANG S， DAI Y， et al. Graph neural networks in node classification： survey and evaluation ［J］. Machine Vision and Applications， 2022， 33（1）： No.4.
[27]	ROSSI A， BARBOSA D， FIRMANI D， et al. Knowledge graph embedding for link prediction： a comparative analysis ［J］. ACM Transactions on Knowledge Discovery from Data， 2021， 15（2）： No.14.
[28]	LUO L， HAFFARI G， PAN S. Graph Sequential Neural ODE Process for link prediction on dynamic and sparse graphs ［C］// Proceedings of the 16th ACM International Conference on Web Search and Data Mining. New York： ACM， 2023： 778-786.
[29]	TAN Y， LIU Y， LONG G， et al. Federated learning on non-IID graphs via structural knowledge sharing ［C］// Proceedings of the 37th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2023： 9953-9961.
[30]	CHEN J， XIONG H， ZHENG H， et al. EGC²： enhanced graph classification with easy graph compression ［J］. Information Sciences， 2023， 629： 376-397.
[31]	JIN W， DERR T， LIU H， et al. Self-supervised learning on graphs： deep insights and new direction ［EB/OL］. ［2024-12-20］. .
[32]	XIAO T， ZHU H， CHEN Z， et al. Simple and asymmetric graph contrastive learning without augmentations ［C］// Proceedings of the 37th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2023： 16129-16152.
[33]	WAN G， TIAN Y， HUANG W， et al. S3GCL： spectral， swift， spatial graph contrastive learning ［C］// Proceedings of the 41st International Conference on Machine Learning. New York： JMLR.org， 2024： 49973-49990.
[34]	KIPF T N， WELLING M. Variational graph auto-encoders ［EB/OL］. ［2024-12-20］. .
[35]	PARK H， LEE S， KIM S， et al. Metropolis-Hastings data augmentation for graph neural networks ［C］// Proceedings of the 35th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2021： 19010-19020.
[36]	SURESH S， LI P， HAO C， et al. Adversarial graph augmentation to improve graph contrastive learning ［C］// Proceedings of the 35th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2021： 15920-15933.
[37]	BERTON L， VALVERDE-REBAZA J， DE ANDRADE LOPES A. Link prediction in graph construction for supervised and semi-supervised learning ［C］// Proceedings of the 2015 International Joint Conference on Neural Networks. Piscataway： IEEE， 2015： 1-8.
[38]	BO D， HU B， WANG X， et al. Regularizing graph neural networks via consistency-diversity graph augmentations ［C］// Proceedings of the 36th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2022： 3913-3921.
[39]	FENG S， JING B， ZHU Y， et al. ArieL： adversarial graph contrastive learning ［J］. ACM Transactions on Knowledge Discovery from Data， 2024， 18（4）： No.82.
[40]	刘渊，杨凯，苏嘉良，等.基于图注意力网络的在线社交网络链路预测［J］.扬州大学学报（自然科学版）， 2024， 27（2）： 21-25.
	LIU Y， YANG K， SU J L， et al. Link prediction for online social networks via graph attention network ［J］. Journal of Yangzhou University （Natural Science Edition）， 2024， 27（2）： 21-25.
[41]	WU S， SUN F， ZHANG W， et al. Graph neural networks in recommender systems： a survey ［J］. ACM Computing Surveys， 2023， 55（5）： No.97.
[42]	VELIČKOVIĆ P， FEDUS W， HAMILTON W L， et al. Deep Graph Infomax ［EB/OL］. ［2024-12-20］. .
[43]	PENG Z， HUANG W， LUO M， et al. Graph representation learning via graphical mutual information maximization ［C］// Proceedings of the Web Conference 2020. New York： ACM， 2020： 259-270.
[44]	CHEN J， KOU G. Attribute and structure preserving graph contrastive learning ［C］// Proceedings of the 37th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2023： 7024-7032.
[45]	JU W， GU Y， MAO Z， et al. GPS： graph contrastive learning via multi-scale augmented views from adversarial pooling ［J］. SCIENCE CHINA Information Sciences， 2025， 68（1）： No.112101.
[46]	GHAYEKHLOO M， NICKABADI A. Supervised contrastive learning for graph representation enhancement ［J］. Neurocomputing， 2024， 588： No.127710.
[47]	SHAO P， TAO J. Multi-level graph contrastive learning ［J］. Neurocomputing， 2024， 570： No.127101.
[48]	WANG H， WEI Z， GAN J， et al. Personalized PageRank to a target node， revisited ［C］// Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York： ACM， 2020： 657-667.
[49]	ZHENG Y， PAN S， LEE V C S， et al. Rethinking and scaling up graph contrastive learning： an extremely efficient approach with group discrimination ［C］// Proceedings of the 36th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2022： 10809-10820.
[50]	HAO Z， XIN H， WEI L， et al. Towards expansive and adaptive hard negative mining： graph contrastive learning via subspace preserving ［C］// Proceedings of the ACM Web Conference 2024. New York： ACM， 2024： 322-333.
[51]	HAMILTON W L， YING Z， LESKOVEC J. Inductive representation learning on large graphs ［C］// Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2017： 1025-1035.
[52]	WU F， ZHANG T， DE SOUZA A H， et al. Simplifying graph convolutional networks ［C］// Proceedings of the 36th International Conference on Machine Learning. New York： JMLR.org， 2019： 6861-6871.
[53]	XU K， HU W， LESKOVEC J， et al. How powerful are graph neural networks？［EB/OL］. ［2024-12-20］. .
[54]	吴宗航，张东，李冠宇.基于联合自监督学习的多模态融合推荐算法［J］.计算机应用， 2025， 45（6）： 1858-1868.
	WU Z H， ZHANG D， LI G Y. Multimodal fusion recommendation algorithm based on joint self-supervised learning ［J］. Journal of Computer Applications， 2025， 45（6）： 1858-1868.
[55]	ZHAO W， YANG K， DING P， et al. Graph attention contrastive learning with missing modality for multimodal recommendation ［J］. Knowledge-Based Systems， 2025， 311： No.113035.

数据集	节点数	边数	特征数	标签数
Cora	2 708	5 429	1 433	7
Citeseer	3 327	4 732	3 703	6
Pubmed	19 717	44 338	500	3
Cora_ML	2 995	16 316	2 879	7
DBLP	17 716	105 734	1 639	4
Photo	19 717	44 338	500	3

数据集	节点数	边数	特征数	标签数
Cora	2 708	5 429	1 433	7
Citeseer	3 327	4 732	3 703	6
Pubmed	19 717	44 338	500	3
Cora_ML	2 995	16 316	2 879	7
DBLP	17 716	105 734	1 639	4
Photo	19 717	44 338	500	3

算法	可用数据	Cora	Citeseer	Pubmed	Cora_ML	DBLP	Photo
GCN^［8］	X， A， Y	81.7	69.2	76.6	81.7	73.9	84.6
GAT^［9］	X， A， Y	82.6	69.6	77.6	82.1	74.2	85.8
DGI^［42］	X， A	83.0	72.0	77.5	82.5	78.1	77.8
GMI^［43］	X， A	80.9	68.1	77.2	82.1	76.8	76.7
GGD^［49］	X， A	83.4	70.4	79.3	81.8	80.2	75.6
SUBG-CON^［21］	X， A	82.8	72.7	79.8	83.5	80.8	89.7
ASP^［44］	X， A	81.8	71.4	79.8	82.7	81.2	91.1
GRAPE^［50］	X， A	84.9	73.1	79.9	86.8	82.6	92.1
SCLDA	X， A	85.3	73.6	80.1	87.5	83.7	92.5

算法	可用数据	Cora	Citeseer	Pubmed	Cora_ML	DBLP	Photo
GCN^［8］	X， A， Y	81.7	69.2	76.6	81.7	73.9	84.6
GAT^［9］	X， A， Y	82.6	69.6	77.6	82.1	74.2	85.8
DGI^［42］	X， A	83.0	72.0	77.5	82.5	78.1	77.8
GMI^［43］	X， A	80.9	68.1	77.2	82.1	76.8	76.7
GGD^［49］	X， A	83.4	70.4	79.3	81.8	80.2	75.6
SUBG-CON^［21］	X， A	82.8	72.7	79.8	83.5	80.8	89.7
ASP^［44］	X， A	81.8	71.4	79.8	82.7	81.2	91.1
GRAPE^［50］	X， A	84.9	73.1	79.9	86.8	82.6	92.1
SCLDA	X， A	85.3	73.6	80.1	87.5	83.7	92.5

编码器	Cora	Citeseer	Pubmed	Cora_ML	DBLP	Photo
GAT^［9］	81.0	72.8	73.9	82.2	81.2	92.2
GraphSage^［51］	81.4	71.7	80.3	85.1	81.5	85.0
SGC^［52］	82.2	73.4	72.1	85.6	82.8	93.1
GIN^［53］	80.8	70.3	68.1	84.3	81.7	90.2
GCN^［8］	85.3	73.6	80.1	87.5	83.7	92.5