首页 | 本学科首页   官方微博 | 高级检索  
     检索      

基于CNN-BiLSTM的消费者网络评论情感分析
引用本文:国显达,那日萨,崔少泽.基于CNN-BiLSTM的消费者网络评论情感分析[J].系统工程理论与实践,2020,40(3):653-663.
作者姓名:国显达  那日萨  崔少泽
作者单位:1. 大连理工大学 系统工程研究所, 大连 116024;2. 大连理工大学 信息与决策技术研究所, 大连 116024
基金项目:国家自然科学基金(61471083);教育部人文社科研究规划基金(14YJA630044)
摘    要:现如今,商品在线评论的情感分析业已成为许多商家不可忽视的重要工作,它对于商家了解用户偏好有着重要意义,同时,它还能够为相关产品下一步的改进工作提供方向指导.然而,传统的分析方法已无法解决现在情感分析中特征提取及语义理解等方面存在的问题.针对此类问题,本文提出一种基于CNN-BiLSTM的在线评论情感分析方法,不仅可以像LSTM一样建立时序关系,而且可以像CNN一样刻画局部空间特征.医疗服务、物流快递、金融服务、旅游住宿和食品餐饮数据集的实验结果表明,该方法能有效判别消费者在线评论情感倾向,在文本的情感分类中效果较传统机器学习算法更准确,F1值可以达到94.67%.同时,实验证明该方法具有较好的领域可拓展性.

关 键 词:情感分析  在线评论  深度学习  CNN模型  BiLSTM模型
收稿时间:2018-09-28

Consumer reviews sentiment analysis based on CNN-BiLSTM
GUO Xianda,ZHAO Narisa,CUI Shaoze.Consumer reviews sentiment analysis based on CNN-BiLSTM[J].Systems Engineering —Theory & Practice,2020,40(3):653-663.
Authors:GUO Xianda  ZHAO Narisa  CUI Shaoze
Institution:1. Institute of Systems Engineering, Dalian University of Technology, Dalian 116024, China;2. Institute of Information and Decision Technology, Dalian University of Technology, Dalian 116024, China
Abstract:Nowadays, the sentiment analysis of online reviews of goods has become an important work that many businesses cannot ignore, which is of great significance for businesses to understand user preferences and can also provide directional guidance for the next improvement of related products. However, the traditional analysis methods have been unable to solve the problems of feature extraction and semantics understanding in sentiment analysis. Aiming at such problems, this paper proposes sentiment analysis of online reviews method based on CNN-BiLSTM, which can not only establish the sequential relationship like LSTM, but also describe local spatial characteristics like CNN. The experimental results on data sets of medical services, logistics express, financial services, tourist accommodation and food and beverage show that this method can effectively discriminate the emotional tendency of consumers' online reviews and is more accurate than traditional machine learning algorithm in sentiment classification in this paper and the F1 value can reach 94.67%. Moreover, the experiments show that the method has good fields expansibility.
Keywords:sentiment analysis  online reviews  deep learning  CNN model  BiLSTM model  
本文献已被 CNKI 维普 等数据库收录!
点击此处可从《系统工程理论与实践》浏览原始摘要信息
点击此处可从《系统工程理论与实践》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号