基于GCN的节点分类研究  

Research on node classification based on GCN

作　　者：张博 宋淑彩[1] 赵一航 ZHANG Bo;SONG Shu-cai;ZHAO Yi-hang(Hebei Institute of Architecture and Civil Engineering,Zhangjiakou,Hebei 075000)

机构地区：[1]河北建筑工程学院,河北张家口075000

出　　处：《河北建筑工程学院学报》2022年第2期196-200,共5页Journal of Hebei Institute of Architecture and Civil Engineering

摘　　要：图数据作为一种非欧几里得数据,因为其数据不具有平移不变性,从而导致常规的神经网络无法很好的对图数据进行特征抽取。但是在最近几年里,深度学习的发展越来越快,非欧几里得的数据也和深度学习组合在了一起,并加入了卷积的技术,形成了图卷积神经网络(Graph Convolutionnal Network),简称GCN。为了更好的从图数据中抽取特征,并准确的对图中没有标签的节点进行预测分类,利用了GCN对Cora数据集进行数据特征信息的抽取,再对提取到的信息进行整合分类,通过对代码当中的学习率、衰减权重和训练次数进行修改调试,最终使准确率从80.5%上升到81.3%。As a kind of non-Euclidean data,graph data does not have translation invariance,which leads to the inability of conventional neural network to extract graph data well.However,in recent years,the development of deep learning is getting faster and faster.Non-euclide data is also combined with deep learning,and convolution technology is added to form graph convolution neural network(GCN for short).In order to extract features from figure data,and accurate forecast no label node in figure classification,Cora data sets to make use of the GCN data characteristic information extraction,and then to extract the information integration of classification and code of each parameter is modified by the,eventually make accuracy increased from 80.5%to 81.3%.

关 键 词：GCN 节点分类 Cora数据 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]