结合图神经网络和图对比学习的半监督多图分类

doi:10.3778/j.issn.1002-8331.2308-0343

摘要/Abstract

摘要： 多图（multi-graph，MG）是一种图袋表示模型，半监督多图分类旨在从有标记和未标记的多图中构建一个预测模型，通过高准确度预测未标记多图，在用户产品推荐、生物制药等领域有着广泛应用。现有基于机器学习的半监督多图分类主要存在两点不足：（1）不能进行全自动的特征选择，过于依赖参数选择。（2）对未标记多图数据的价值未充分挖掘。因此，提出一种结合图神经网络和图对比学习的半监督多图分类方法（graph neural network combining with graph contrastive learning for semi-supervised multi-graph classification，GCSS）。一方面，分别设计从局部和全局提取特征信息的模块，并引入NN协同器（neural networks collaborator，NN collaborator）完成这两个模块的协作，自适应学习数据的特征表示进行训练；另一方面，采用图对比学习（graph contrastive learning，GCL）和半监督学习（semi-supervised learning，SSL）从两个不同学习视角来充分利用未标记多图数据，降低模型对标签等的依赖。在真实数据集上的大量实验结果验证了所提出方法的预测性能均优于基线方法。

关键词: 半监督多图分类, 图对比学习, 图神经网络, 注意力机制

Abstract: Multi-graph (MG) is a representation model of the bag-of-graphs, semi-supervised multi-graph classification aims to build a prediction model from marked and unmarked multi-graphs. Through high-accuracy prediction of unmarked multi-graphs, it’s widely used in user product recommendation, biopharmaceuticals and other fields. There are two main shortcomings in the existing semi-supervised multi-graph classification based on machine learning: (1) It’s cannot be fully automatic feature selection and relies too much on parameter selection. (2) The value of unmarked multi-graph data is not fully mined. Therefore, a graph neural network combining with graph contrastive learning for semi-supervised multi-graph classification method (GCSS) is proposed. On the one hand, it designs modules that extract feature information from local and global respectively, and introduces neural networks collaborator (NN collaborator) to complete the collaboration of these two modules, and trains the feature representation of adaptive learning data. On the other hand, graph contrastive learning (GCL) and semi-supervised learning (SSL) are used to make full use of the unmarked from two different learning perspectives, it reduces the model’s dependence on labels, etc. A large number of experimental results on the real dataset verify that the prediction performance of the proposed method is better than that of the baseline method.

Key words: semi-supervised multi-graph classification, graph contrastive learning, graph neural network, attention mechanism

路秋霖, 王慧颖, 朱峰冉, 李全鑫, 庞俊. 结合图神经网络和图对比学习的半监督多图分类[J]. 计算机工程与应用, 2025, 61(1): 368-374.

LU Qiulin, WANG Huiying, ZHU Fengran, LI Quanxin, PANG Jun. Semi-Supervised Multi-Graph Classification Combining Graph Neural Network and Graph Contrastive Learning[J]. Computer Engineering and Applications, 2025, 61(1): 368-374.

参考文献

[1] ZHANG D, YIN J, ZHU X, et al. Network representation learning: a survey[J]. IEEE Transactions on Big Data, 2020, 6(1): 3-28.
[2] WU J, ZHU X, ZHANG C, et al. Bag constrained structure pattern mining for multi-graph classification[J]. IEEE Transactions on Knowledge and Data Engineering, 2014, 26(10): 2382-2396.
[3] JIA L. Semi-supervised multi-class classification algorithm based on local learning[M]//Information engineering and applications. London: Springer, 2012: 111-118.
[4] ZHI H Y, ZHAO L, LEE C C, et al. A novel graph neural network methodology to investigate dihydroorotate dehydrogenase inhibitors in small cell lung cancer[J]. Biomolecules, 2021, 11(3): 477.
[5] PANG J, GU Y, XU J, et al. Semi-supervised multi-graph classification using optimal feature selection and extreme learning machine[J]. Neurocomputing, 2018, 277: 89-100.
[6] YANG G, ZHANG N, JIN T, et al. Research on semi-supervised multi-graph classification algorithm based on MR-MGSSL for sensor network[J]. EURASIP Journal on Wireless Communications and Networking, 2020, 2020(1): 130.
[7] WANG R, WANG X, SHI C, et al. Uncovering the structural fairness in graph contrastive learning[J]. arXiv:2210.03011, 2022.
[8] 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]//Advances in Neural Information Processing Systems, 2022, 35: 10809-10820.
[9] YOU Y, CHEN T, SUI Y, et al. Graph contrastive learning with augmentations[C]//Advances in Neural Information Processing Systems, 2020, 33: 5812-5823.
[10] VELICKOVIC P, FEDUS W, HAMILTON W L, et al. Deep graph infomax[J]. arXiv:1809.10341, 2018
[11] 王永贵, 赵晓暄. 结合自监督学习的图神经网络会话推荐[J]. 计算机工程与应用, 2023, 59(3): 244-252.
WANG Y G, ZHAO X X. Self-supervised graph neural networks for session-based recommendation[J]. Computer Engineering and Applications, 2023, 59(3): 244-252.
[12] 蒋光峰, 胡鹏程, 叶桦, 等. 基于重构误差的同构图分类模型[J]. 计算机科学与探索, 2022, 16(1): 185-193.
JIANG G F, HU P C, YE H, et al. Isomorphic graph classification model based on reconstruction error[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(1): 185-193.
[13] YING C, CAI T, LUO S, et al. Do transformers really perform badly for graph representation[C]//Advances in Neural Information Processing Systems, 2021, 34: 28877-28888.
[14] WU J, PAN S, ZHU X, et al. Positive and unlabeled multi-graph learning[J]. IEEE Transactions on Cybernetics, 2017, 47(4): 818-829.
[15] ACHANTA R, SHAJI A, SMITH K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(11): 2274-2282.
[16] XU K, HU W, LESKOVEC J, et al. How powerful are graph neural networks?[J]. arXiv:1810.00826, 2018.