摘要
目前,微博情感分析方法存在以下2方面问题:1)对于微博文本的情感语义表示模型存在缺陷,不能有效表示文本整体的情感语义信息.2)多采用全局分类器,对于细粒度情感分析,往往效果不佳.针对以上问题,本文提出一种BGRU-CNN神经网络模型,结合基于双向门控循环单元的神经网络和卷积神经网络来训练分类器,并采用层次结构分类方法进一步提高了模型在细粒度微博情感分类任务上的效果.在NLPCC2014微博情感分析数据集上进行实验,取得了比传统模型和方法更好的分类效果.
At present,there are two problems in micro-blog sentiment analysis: 1) There are defects in the emotional semantic representation model of micro-blog text,which can not effectively express the emotional semantic information of the whole text. 2) Most of them use global classifier,which is not effective for fine-grained sentiment analysis. To address these problems,we proposed a BGRU-CNN neural network model,which combines the bidirectional gated recurrent units network with the convolutional neural network to train the classifier. And a hierarchical structure classification method is used to further improve the effectiveness in the fine grained micro-blog sentiment classification task. The experiment on NLPCC2014 microblog sentiment analysis dataset has achieved better classification results than that with traditional models and methods.
作者
刘高军
赵希明
LIU Gaojun;ZHAO Ximing(Col.of Information,North China Univ.of Tech.,100144,Beijing,China)
出处
《北方工业大学学报》
2019年第2期68-76,共9页
Journal of North China University of Technology
基金
新闻出版业科技与标准重点实验室——CNONIX国家标准应用与推广实验室“CNONIX数据符合性测试”项目(4020548418H1)
关键词
情感分析
循环神经网络
卷积神经网络
BGRU-CNN
层次结构
sentiment analysis
recurrent neural network
convolutional neural network
BGRU-CNN
hierarchical structure