文本特征和图结点混合增强的图卷积网络文本分类
Hybrid Augmentation of Text Feature and Graph Node for Graph Convolutional Networks Text Classification
杨晓奇 1刘伍颖2
作者信息
- 1. 广东外语外贸大学信息科学与技术学院,广东广州 510006
- 2. 鲁东大学山东省语言资源开发与应用重点实验室,山东烟台 264025;广东外语外贸大学外国语言学及应用语言学研究中心,广东广州 510420
- 折叠
摘要
在BertGCN模型的基础上改进其结构,同时结合文本特征和图结点混合增强的方法,使用新的边权重计算算法BM25+构造图的边.使用R8、R52、Ohsumed和MR这4个常用的公开数据集来验证所提方法的有效性.结果表明:与BertGCN模型及其它基线模型相比,该方法在4个文本分类数据集上的准确率评价指标均有不同程度的提升.
Abstract
The work will improve the structure on the basis of the BertGCN model,not only using a new algorithm to construct the edges of the graph,but also combining a hybrid enhancement of text features and graph nodes.The method not only has some optimization in the edge structure,but also makes fuller use of the extended semantic information of the text in the form of text feature enhancement and graph-enhanced nodes,while retaining the original text features.Four public datasets,R8,R52,Ohsumed and MR which are commonly used,are used to verify the effectiveness of this method.The experimental results show that compared with the BertGCN model and other baselines,the accuracy evaluation metric of the method on the four text classification data sets has been improved to varying degrees.
关键词
BM25+/文本特征增强/图结点增强/预训练模型/图卷积网络/文本分类Key words
BM25+/text feature enhancement/graph node enhancement/pre-training model/graph convolution networks/text classification引用本文复制引用
基金项目
教育部人文社会科学研究规划基金(20YJAZH069)
&&(20YJC740062)
上海市哲学社会科学规划项目(2019BYY028)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据