一种基于CNN与双向LSTM融合的文本情感分类方法  

A text emotion classification method based on CNN and bidirectional LSTM fusion

作　　者：张翠[1] 周茂杰[1] Zhang Cui;Zhou Maojie(BowenCollege of Management,Guilin University of Technology,Guilin,Guangxi 541006,China;Guilin University of Technology)

机构地区：[1]桂林理工大学博文管理学院

出　　处：《计算机时代》2019年第12期38-41,共4页Computer Era

基　　金：“广西高校中青年教师基础能力提升项目”(项目编号：2018KY0854);广西民办高校重点支持建设专业项目的支持

摘　　要：现在文本情感分类普遍采用深度学习的方法。卷积神经网络可以较好地提取局部特征,但是缺少对上下文的理解。长短记忆网络可以有效记忆较长距离的信息,有较强的全局性。为实现全局特征与局部特征的有效融合,研究了一种融合两种特征的深度学习方法,构建深度学习网络模型。利用互联网中获取的文本作为训练语料及测试语料,在百度开源平台PaddlePaddle上进行实验。实验结果显示,该算法与传统CNN和LSTM模型算法相比,识别的准确率分别提高了2.65和1.87个百分点,说明该模型算法在文本情感分类的性能上有所提高。Nowadays,deep learning is widely used in text emotion classification.CNN(convolutional neural networks)can extract local features well,but it lacks the understanding of context.LSTM(long and short memory networks)can effectively memorize long-distance information and have a strong global character.In order to achieve the effective integration of global and local features,this paper studies a deep learning method with the fusion of the two,and constructs a deep learning network model.Using the text obtained from the Internet as training corpus and testing corpus,experiments are carried out on Baidu open source platform PaddlePaddle.The experiment results show that compared with the traditional CNN and LSTM models,the recognition accuracy of the proposed algorithm is improved by 2.65 and 1.87 percentage points respectively,which shows that the performance of this model in text emotion classification is improved.

关 键 词：卷积神经网络 双向长短记忆网络 融合 情感分类 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]