一种融合图数据多元结构和特征的图池化方法

doi:10.19678/j.issn.1000-3428.0068357

摘要/Abstract

摘要：

在图神经网络中, 图池化是一类用于对图数据进行下采样以提取图表征的重要操作。由于图数据存在较为复杂的网络拓扑结构和高维度的特征信息, 因此现有图池化方法在设计过程中未能同时融合图数据的拓扑结构信息和节点的长距离依赖信息, 在图池化过程中没有考虑丢弃节点的特征, 造成图数据的重要信息损失。为此, 提出一种基于多元特征融合的图池化方法来同时捕获图数据的局部拓扑信息、全局拓扑信息以及长距离节点依赖关系, 并使用1个聚合模块聚合这些特征信息得到1个新的池化图。为了解决图池化过程中节点特征信息丢失的问题, 提出一种新的特征融合方法将丢弃节点的信息以一定比例汇聚到保留节点上。基于该池化方法, 构建基于分层池化的图分类模型。在D&D、PROTEINS、NCI1和NCI109 4个数据集上的实验结果表明, 与最佳基线模型相比, 所提模型的分类准确率分别提升了2.97、3.59、0.48和0.24个百分点, 能够更有效利用图数据的特征信息、拓扑信息和长距离节点依赖信息, 在图分类任务上取得了更好的效果。

关键词: 图池化, 图分类, 拓扑信息, 长距离节点依赖, 特征融合

Abstract:

In graph neural networks, graph pooling is a critical operation used to downsample graph data and extract graph representations. Owing to the complex network topology and high-dimensional feature information of graph data, existing graph pooling methods fail to simultaneously integrate both the topological information of graph data and the long-distance dependency information of nodes during the design process. In the graph pooling process, node features are not discarded because discarding them would result in the loss of important information from the graph data. To address these issues, this study proposes a graph pooling method based on multi-feature fusion to simultaneously capture the local and global topology structures and long-distance dependencies of graph data. An aggregation module is then used to combine these features to obtain a new pooled graph. To solve the problem of node feature information loss during graph pooling, a new feature fusion method is proposed to aggregate the information of discarded nodes in a certain proportion onto the reserved nodes. Using this pooling method, a graph classification model is constructed based on hierarchical pooling. The experimental results on four datasets-D&D, PROTEINS, NCI1, and NCI109-indicate that compared with the best baseline model, the proposed model improves the classification accuracy by 2.97, 3.59, 0.48, and 0.24 percentage points, respectively. It can more effectively utilize the features, topological, and long-distance node-dependency information of graph data, and achieve better results in graph classification tasks.

Key words: graph pooling, graph classification, topological information, long-distance node dependencies, feature fusion

王翔, 魏玉锌, 毛国君. 一种融合图数据多元结构和特征的图池化方法[J]. 计算机工程, 2025, 51(1): 128-137.

WANG Xiang, WEI Yuxin, MAO Guojun. A Graph Pooling Method Fusing Multiple Structures and Features of Graph Data[J]. Computer Engineering, 2025, 51(1): 128-137.

https://www.ecice06.com/CN/Y2025/V51/I1/128

图/表 14

图1 MFFPool模型结构

Fig.1 Structure of MFFPool model

图2 基于多头自注意力机制的池化模块结构

Fig.2 Structure of pooling module based on multi-head self-attention mechanism

图3 基于节点聚类的池化模块结构

Fig.3 Structure of pooling module based on node clustering

图4 基于图卷积的池化模块结构

Fig.4 Structure of pooling module based on graph convolution

图5 图分类模型结构

Fig.5 Structure of graph classification model

图6 不同特征重聚合比例对准确率的影响

Fig.6 The impact of different feature reaggregation ratios on accuracy

图7 不同变体与所提模型的准确率对比

Fig.7 Comparison of accuracy between different variants and the proposed model

图8 不同模型生成的图嵌入的t-SNE可视化

Fig.8 t-SNE visualization of graph embeddings generated by different models

图9 池化比例对模型准确率的影响

Fig.9 The influence of pooling ratio on accuracy of model

图10 不同网络层数下的模型准确率

Fig.10 Accuracy of model under different number of network layers

图11 多头自注意力机制头数对模型准确率的影响

Fig.11 The influence of head number of self-attention on model accuracy

参考文献 26

1	程章桃, 钟婷, 张晟铭, 等. 基于图学习的推荐系统研究综述. 计算机科学, 2022, 49 (9): 1- 13.
	CHENG Z T , ZHONG T , ZHANG S M , et al. Survey of recommender systems based on graph learning. Computer Science, 2022, 49 (9): 1- 13.
2	HASAN M, ZAKI M J. A survey of link prediction in social networks[M]//Social Network Data Analytics. Berlin, Germany: Springer, 2011: 243-275.
3	王兆慧, 沈华伟, 曹婍, 等. 图分类研究综述. 软件学报, 2022, 33 (1): 171- 192.
	WANG Z H , SHEN H W , CAO Q , et al. Survey on graph classification. Journal of Software, 2022, 33 (1): 171- 192.
4	KIPF T N, WELLING M. Semi-supervised classification with graph convolutional networks[EB/OL]. [2023-08-01]. https://arxiv.org/pdf/1609.02907.
5	XU K, HU W H, LESKOVEC J, et al. How powerful are graph neural networks?[EB/OL]. [2023-08-01]. https://arxiv.org/pdf/1810.00826.
6	马涪元, 王英, 李丽娜, 等. 融合结构和特征的图层次化池化模型. 计算机科学与探索, 2023, 17 (1): 179- 186.
	MA F Y , WANG Y , LI L N , et al. Structure and feature fusion graph hierarchical pooling model. Journal of Frontiers of Computer Science and Technology, 2023, 17 (1): 179- 186.
7	YING R, YOU J X, MORRIS C, et al. Hierarchical graph representation learning with differentiable pooling[EB/OL]. [2023-08-01]. https://arxiv.org/pdf/1806.08804.
8	徐立祥, 葛伟, 陈恩红, 等. 基于图核同构网络的图分类方法. 计算机研究与发展, 2024, 61 (4): 903- 915.
	XU L X , GE W , CHEN E H , et al. Graph classification method based on graph kernel isomorphism network. Journal of Computer Research and Development, 2024, 61 (4): 903- 915.
9	LEE J, LEE I, KANG J. Self-attention graph pooling[EB/OL]. [2023-08-01]. https://arxiv.org/abs/1904.08082v4.
10	HUANG J J, LI Z H, LI N N, et al. AttPool: towards hierarchical feature representation in graph convolutional networks via attention mechanism[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Washington D. C., USA: IEEE Press, 2019: 6479-6488.
11	GAO X , DAI W R , LI C L , et al. iPool—information-based pooling in hierarchical graph neural networks. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33 (9): 5032- 5044. doi: 10.1109/TNNLS.2021.3067441
12	BRUNA J, ZAREMBA W, SZLAM A, et al. Spectral networks and deep locally connected networks on graphs[EB/OL]. [2023-08-01]. https://arxiv.org/pdf/1312.6203v2.
13	DEFFERRARD M, BRESSON X, VANDERGHEYNST P. Convolutional neural networks on graphs with fastlocalized spectral filtering[EB/OL]. [2023-08-01]. https://arxiv.org/pdf/1606.09375.
14	XU B B, SHEN H W, CAO Q, et al. Graph wavelet neural network[EB/OL]. [2023-08-01]. https://arxiv.org/abs/1904.07785v1.
15	HAMILTON W L, YING R, LESKOVEC J. Inductive representation learning on large graphs[EB/OL]. [2023-08-01]. https://arxiv.org/pdf/1706.02216.
16	VELICKOVIC P, CUCURULL G, CASANOVA A, et al. Graph attention networks[EB/OL]. [2023-08-01]. https://arxiv.org/pdf/1710.10903.
17	DU J L, WANG S Z, MIAO H, et al. Multi-channel pooling graph neural networks[C]//Proceedings of the 13th International Joint Conference on Artificial Intelligence. Montreal, Canada: International Joint Conferences on Artificial Intelligence Organization, 2021: 1442-1448.
18	VINYALS O, BENGIO S, KUDLUR M. Order matters: Sequence to sequence for sets[EB/OL]. [2023-08-01]. https://arxiv.org/pdf/1511.06391.
19	SHI X J, CHEN Z R, WANG H, et al. Convolutional LSTM network: a machine learning approach for precipitation nowcasting[C]//Proceedings of the 28th International Conference on Neural Information Processing Systems. New York, USA: ACM Press, 2015: 802-810.
20	ZHANG M H, CUI Z C, NEUMANN M, et al. An end-to-end deep learning architecture for graph classification[C]//Proceedings of the AAAI Conference on Artificial Intelligence. [S. l.]: AAAI Press, 2018: 4438-4445.
21	MA Y, WANG S H, AGGARWAL C C, et al. Graph convolutional networks with eigenpooling[C]//Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York, USA: ACM Press, 2019: 723-731.
22	GAO H Y, JI S W. Graph U-Nets[EB/OL]. [2023-08-01]. https://arxiv.org/abs/1905.05178.
23	DOBSON P D , DOIG A J . Distinguishing enzyme structures from non-enzymes without alignments. Journal of Molecular Biology, 2003, 330 (4): 771- 783. doi: 10.1016/S0022-2836(03)00628-4
24	BORGWARDT K M, ONG C S, SCHÖNAUER S, et al. Protein function prediction via graph kernels[C]//Proceedings the 13th International Conference on Intelligent Systems for Molecular Biology. New York, USA: ACM Press, 2005: 47-56.
25	WEI L N, HE Z Q, ZHAO H, et al. Search to capture long-range dependency with stacking GNNs for graph classification[C]//Proceedings of the ACM Web Conference. New York, USA: ACM Press, 2023: 588-598.
26	DUAN Y T , WANG J M , MA H R , et al. Residual convolutional graph neural network with subgraph attention pooling. Tsinghua Science and Technology, 2022, 27 (4): 653- 663. doi: 10.26599/TST.2021.9010058

[1]	周宇, 谢威, 邝得互, 江健民. 基于三元自注意力的视频快照压缩成像重建[J]. 计算机工程, 2025, 51(1): 20-30.
[2]	费涛, 艾山·吾买尔, 杜文旭, 朱翠翠. 基于Squeezeformer的多颗粒度多方面发音质量评测方法[J]. 计算机工程, 2025, 51(1): 81-87.
[3]	胡涌涛, 黄洪琼. 结合特征融合和通道注意力的多分支换装行人重识别[J]. 计算机工程, 2025, 51(1): 225-234.
[4]	李猛坤, 袁晨, 王琪, 赵冲, 陈景轩, 刘立峰. 基于改进YOLOv8算法的在线听课行为识别模型研究[J]. 计算机工程, 2025, 51(1): 287-294.
[5]	李俊仪, 李向阳, 龙朝勋, 李海燕, 李红松, 余鹏飞. 基于多级区域选择与跨层特征融合的野生菌分类[J]. 计算机工程, 2024, 50(9): 179-188.
[6]	张华青, 夏张涛, 陆晓庆, 童基均. 基于字形特征的血管外科命名实体识别[J]. 计算机工程, 2024, 50(8): 13-21.
[7]	李华昱, 张智康, 闫阳, 岳阳. 基于知识图谱增强的领域多模态实体识别[J]. 计算机工程, 2024, 50(8): 31-39.
[8]	陈瀚, 赵春蕾, 蒋昊达, 王春东. 基于融合模型与语义网络的App用户意图识别研究[J]. 计算机工程, 2024, 50(8): 50-63.
[9]	刘锁兰, 王炎, 王洪元, 朱生升. 基于多流语义图卷积网络的人体行为识别[J]. 计算机工程, 2024, 50(8): 64-74.
[10]	赵婉秋, 张俊虎, 李海涛. 用于建筑物分割的平行结构特征融合网络[J]. 计算机工程, 2024, 50(8): 239-248.
[11]	赵宏, 王枭. 基于Swin-Transformer的黑色素瘤图像病灶分割研究[J]. 计算机工程, 2024, 50(8): 249-258.
[12]	王富平, 刘鸿玮, 张锲石, 段冠庄. 基于深度特征抑制的遮挡人脸识别网络[J]. 计算机工程, 2024, 50(8): 259-269.
[13]	闵莉, 董冰洁, 安冬. 基于多注意力机制与跨特征融合的语义分割算法[J]. 计算机工程, 2024, 50(8): 282-289.
[14]	陈宇航, 杨勇, 先木斯亚·买买提明, 帕力旦·吐尔逊, 樊小超, 任鸽, 刁宇峰. 基于主题感知和语义增强的作文自动评分方法[J]. 计算机工程, 2024, 50(8): 363-371.
[15]	谭巨全, 王然. 特征融合下田径录像3D人体动作DTW捕捉算法[J]. 计算机工程, 2024, 50(7): 71-78.

选择文件类型/文献管理软件名称

选择包含的内容