摘要
图池化作为图神经网络中重要的组件,在获取图的多粒度信息的过程中扮演了重要角色。而当前的图池化操作均以平等地位看待数据点,普遍未考虑利用邻域内数据之间的偏序关系,从而造成图结构信息破坏。针对此问题,本文提出一种基于偏序关系的多视图多粒度图表示学习框架(multi-view and multi-granularity graph representation learning based on partial order relationships,MVMGr-PO),它通过从节点特征视图、图结构视图以及全局视图对节点进行综合评分,进而基于节点之间的偏序关系进行下采样操作。相比于其他图表示学习方法,MVMGr-PO可以有效地提取多粒度图结构信息,从而可以更全面地表征图的内在结构和属性。此外,MVMGr-PO可以集成多种图神经网络架构,包括GCN(graph convolutional network)、GAT(graph attention network)以及GraphSAGE(graph sample and aggregate)等。通过在6个数据集上进行实验评估,与现有基线模型相比,MVMGr-PO在分类准确率上有明显提升。
Graph pooling,as a crucial component of graph neural networks(GNNs),plays a vital role in capturing multigranularity information of graphs.However,current graph pooling operations typically treat data points equally,often neglecting the partial order relationships among data within neighborhoods,which leads to the disruption of graph structural information.To address this issue,we propose a novel framework for multi-view and multi-granularity graph representation learning based on partial order relationships,named MVMGr-PO.This framework comprehensively scores nodes from the perspectives of node feature view,graph structure view,and global view,and then performs downsampling operations based on the partial order relationships among nodes.Compared with other graph representation learning methods,MVMGr-PO effectively extracts multi-granularity graph structural information,thus providing a more comprehensive representation of the intrinsic structure and attributes of the graph.Additionally,MVMGr-PO can integrate various graph neural network(GNN)architectures,including graph convolutional network(GCN),graph attention network(GAT),and graph sample and aggregate(GraphSAGE).Experimental evaluations on six datasets demonstrate that compared with existing baseline models,MVMGr-PO significantly improves classification accuracy.
作者
肖添龙
徐计
王国胤
XIAO Tianlong;XU Ji;WANG Guoyin(State Key Laboratory of Public Big Data,Guizhou University,Guiyang 550025,China;College of Computer Science and Technology,Guizhou University,Guiyang 550025,China;Chongqing Key Laboratory of Computational Intelligence,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)
出处
《智能系统学报》
北大核心
2025年第1期243-254,共12页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金项目(62366008,61966005,62221005).
关键词
图神经网络
图池化
多粒度
偏序关系
节点分类任务
图表示学习
半监督学习
图嵌入
graph neural networks
graph pooling
multi-granularity
partial order relationships
node classification task
graph representation learning
semi-supervised learning
graph embedding
