多特征融合的Voting-SRM情感分类研究  

Voting-SRM Sentiment Classification Based on Multi-feature Fusion

作　　者：赵乐 麦范金[1] 张兴旺[2] ZHAO Le;MAI Fan-jin;ZHANG Xing-wang(Guilin University of Technology.School of Information Science and Engineering,Colloge of Mechanical and Control Enineering,Guilin 541006,China;Guilin University of Technology,Library,Guilin 541006,China)

机构地区：[1]桂林理工大学信息科学与工程学院,机械与控制工程学院,广西桂林541006 [2]桂林理工大学图书馆,广西桂林541006

出　　处：《小型微型计算机系统》2019年第11期2269-2273,共5页Journal of Chinese Computer Systems

基　　金：国家社科基金青年项目(17CTQ004)资助

摘　　要：情感分类是自然语言处理领域的一个核心问题,其目的是判断评论文本的情感极性,并挖掘其蕴含的情感价值信息.为了提取评论文本中潜在的情感信息,提高分类精度,本文提出了多特征融合的Voting-SRM情感分类方法.结合词性特征,语法特征等,提取名词,动词,形容词,副词等特征,然后运用软投票机制,结合随机梯度下降算法、随机森林、神经网络等算法,对已获取评论文本进行极性二分类.本文通过对比实验,验证了该方法的有效性.Emotional classification is a core issue in the field of natural language processing. Its purpose is to judge the emotional polarity of the review text and to mine the emotional value information it contains. In order to extract the potential emotional information in the review text and improve the classification accuracy,this paper proposes a multi-feature fusion Voting-SRM sentiment classification method.Combining part-of-speech features,grammatical features,etc.,extract nouns,verbs,adjectives,adverbs and other characteristics,and then combine the soft voting mechanism with random gradient descent,random forest,neural network and other algorithms to classify the obtained review texts. In this paper,the effectiveness of the method is validated by comparison experiments.

关 键 词：词性标注 二元语法 随机梯度下降 投票机制 情感分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]