图神经网络在招投标文件分类中的应用  

Application of Graph Neural Network in the Classification of Bidding Documents

作　　者：强成宇 李晓戈 马鲜艳[1] 李涛 田俊鹏 QIANG Cheng-yu;LI Xiao-ge;MA Xian-yan;LI Tao;TIAN Jun-peng(College of Computer,Xi′an University of Posts and Telecommunications,Xi′an 710000,China;Shannxi Provincial Key Laboratory of Network Data Analysis and Intelligent Processing(Xi′an University of Posts and Telecommunications),Xi′an 710000,China)

机构地区：[1]西安邮电大学计算机学院,西安710000 [2]西安邮电大学陕西省网络数据分析与智能处理重点实验室,西安710000

出　　处：《小型微型计算机系统》2023年第2期275-280,共6页Journal of Chinese Computer Systems

基　　金：国家重点研发计划项目(2018YFB1402905)资助;陕西省重点研发计划项目(2020GY-227)资助.

摘　　要：全国政府机关、事业单位的采购网站每天都会发布数万条招投标信息,如何快速有效的分类这些数据,成为挖掘其相应价值的关键.本文针对网络上招投标文件缺乏标注、文本语义稀疏、数据来源多样、信息结构复杂等问题,提出了一种基于图卷积神经网络的半监督分类方法(BD-GCN).该方法首先将爬取的招投标文件进行结构化清洗,并利用信息抽取技术构建为特殊的知识图谱模型,再融合外部文本信息,最后采用图卷积神经网络实现招投标文件的半监督分类.本文利用在网络上爬取的36123条招投标文件进行实验,并与当前流行的分类方法进行对比.实验结果表明,BD-GCN能有效提高分类的准确率.Procurement websites of government organs and public institutions across the country will publish thousands of bidding information every day.How to classify these data quickly and effectively has become the key to excavate their corresponding value.In order to solve the problems of bidding documents on the Internet,such as lack of annotation,sparse text semantics,diverse data sources and complex information structure,this paper proposes a semi-supervised bidding documents classification method(BD-GCN)based on Graph Convolutional Networks.In this method,the bidding documents crawled were structurally cleaned firstly,and the information extraction technology was used to build a special knowledge graph graph model,and then the external text information was fused.Finally,the Graph Convolutional Network was used to realize the semi-supervised classification of bidding documents.In this paper,36123 tendering and bidding documents crawled from the Internet are used for experiments,and compared with the current popular classification methods.The experimental results show that BD-GCN can effectively improve the accuracy of classification.

关 键 词：文本分类 知识图谱 多源异构图 图卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]