Journal of Frontiers of Computer Science and Technology ›› 2019, Vol. 13 ›› Issue (6): 973-981.DOI: 10.3778/j.issn.1673-9418.1806018

Previous Articles     Next Articles

BGRU: New Method of Chinese Text Sentiment Analysis

CAO Yu, LI Tianrui+, JIA Zhen, YIN Chengfeng   

  1. School of Information Science and Technology, Southwest Jiaotong University, Chengdu 611756, China
  • Online:2019-06-01 Published:2019-06-14

BGRU:中文文本情感分析的新方法

曹  宇,李天瑞+,贾  真,殷成凤   

  1. 西南交通大学 信息科学与技术学院,成都 611756

Abstract: Social network is an indispensable part of social life, and the text sentiment analysis of social network has become a research hot topic of natural language processing. In view that deep learning technology can automatically build text features, people have proposed deep learning models such as CNN (convolutional neural network) and BLSTM (bidirectional long short-term memory) to solve text sentiment analysis, but there are still problems such as complicated structures or long training time. BGRU (bidirectional gated recurrent unit) can remember the context information of sequence, the structure is relatively simple and the training speed is fast. A Chinese text sentiment analysis method based on BGRU is proposed. This method firstly converts the text into a sequence of word vector. Then it obtains the text sentiment features through BGRU. Finally, the classifier identifies the text sentiment orientation. The experimental results on the ChnSentiCorp corpus achieve 90.61% of [F1] value. It is shown that this method is superior to CNN and BLSTM, etc. And training speed is 1.36 times higher than that of BLSTM.

Key words: bidirectional gated recurrent unit (BGRU), deep learning, sentiment analysis

摘要: 社交网络作为社会生活不可或缺的一部分,针对其产生的文本数据进行情感分析已成为自然语言处理领域的一个研究热点。鉴于深度学习技术能够自动构建文本特征,人们已提出CNN(convolutional neural network)、BLSTM(bidirectional long short-term memory)等模型来解决文本情感分析问题,但还存在结构较为复杂或训练时间较长等问题,而BGRU(bidirectional gated recurrent unit)能记忆序列的上下文信息,并且结构较为简单,训练速度较快。提出一种基于BGRU的中文文本情感分析方法,首先将文本转换为词向量序列,然后利用BGRU获得文本的上下文情感特征,最后由分类器给出文本的情感倾向。在ChnSentiCorp语料上进行实验,该方法取得了90.61%的[F1]值,效果优于CNN和BLSTM等模型,并且训练速度是BLSTM的1.36倍。

关键词: 双向门控循环单元(BGRU), 深度学习, 情感分析