基于字符级双通道复合网络的中文文本情感分析  

Chinese text sentiment analysis based on character-level two-channel composite network

作　　者：王丽亚 刘昌辉[1] 蔡敦波[1] 赵彤洲[1] 王梦 Wang Liya;Liu Changhui;Cai Dunbo;Zhao Tongzhou;Wang Meng(College of Computer Science&Engineering,Wuhan Institute of Technology,Wuhan 430205,China)

机构地区：[1]武汉工程大学计算机科学与工程学院,武汉430205

出　　处：《计算机应用研究》2020年第9期2674-2678,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61103136);武汉工程大学教育创新计划资助项目(CX2018196)。

摘　　要：针对传统卷积神经网络(CNN)缺乏句子体系特征的表示,以及传统双向门限循环神经网络(BiGRU)缺乏提取深层次特征能力。以中文文本为研究对象,在字符级词向量的基础上提出双通道的CNN-BiGRU复合网络,同时引入注意力机制的模型进行情感分析。首先,在单通道上利用CNN提取深层次短语特征,利用BiGRU提取全局特征的能力深度学习短语体系特征,从而得到句子体系的特征表示;再通过增加注意力层进行有效特征筛选;最后,采用双通道结构的复合网络,丰富了特征信息,加强了模型的特征学习能力。在数据集上进行多组对比实验,该方法取得了92.73%的F1值结果,优于对照组,说明提出的模型能有效地提高文本分类的准确率。同时在单句测试上量化出模型优势,且实现了模型的实际应用能力。The traditional CNN lacks the representation of sentence system features,and the traditional BiGRU lacks the ability to extract deep features. This paper took the Chinese text as the research object,proposed a two-channel CNN-BiGRU composite network based on the character-level word vector,and introduced the attention mechanism model for sentiment analysis.Firstly,using CNN to extract deep-level phrase features on a single channel,and the ability to extract global features using BiGRU,it studied the characteristics of the phrase system deeply,so as to obtain the feature representation of the sentence system. Then it screened effective feature by increasing the attention layer. Finally,the composite network with two-channel structure enriched the feature information and enhanced the feature learning ability of the model. In the multi-group comparison experiment on the dataset,the method obtained 92. 73% F1 value is better than the control group,indicating that the proposed model can effectively improve the accuracy of text classification.

关 键 词：卷积神经网络 双向门限循环神经网络 注意力机制 中文文本情感分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]