«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

26

期数:

2016年01期

页码:

74-77

栏目:

智能、算法、系统工程

出版日期:

2016-01-10

文章信息/Info

Title:

 Identifying of Word Sentiment Orientation of Transductive Learning and Semantic Comprehension

文章编号:

1673-629X（2016)01-0074-04

作者:

 闻彬; 饶彬; 赵君喆; 焦翠珍; 戴文华

 湖北科技学院 计算机科学与技术学院

Author(s):

 WEN Bin; RAO Bin; ZHAO Jun-zhe; JIAO Cui-zhen; DAI Wen-hua

关键词:

 ; 词语倾向性识别; 机器学习; 语义理解; 意见挖掘; 情感义原; HowNet

Keywords:

 ; word sentiment orientation; machine learning; semantic comprehension; opinion mining; sentimental primitive; HowNet

分类号:

TP391.1

文献标志码:

A

摘要:

 目前词语情感倾向性识别研究主要分为机器学习和语义理解，机器学习不能很好地识别通用领域词语，语义理解又存在准确率和召回率不够高的问题，因此文中提出了一种融合直推式学习和语义理解的词语倾向性识别方法。首先对HowNet 知识库体系进行改进，在已有的四种义原的基础上，提出第五义原—情感义原；然后将第五义原手工融入到 How-Net 知识库中，再在此基础上提出词语情感相似度计算方法计算词语的情感值；最后将该方法融合直推式学习以判定词语情感倾向性。通过实验结果表明，与支持向量机和原语义理解方法相比，该方法在识别情感词上取得了较好的效果。

Abstract:

 At present,the research on word sentiment orientation identification is mainly divided into machine learning and semantic com-prehension,but machine learning cannot handle general field words effectively,semantic comprehension also cannot get high scores at pre-cision and recall,therefore,a new fusion method between transductive learning and semantic comprehension for judging word polarity was put forward in this paper. Firstly the HowNet knowledge base system is improved,on the basis of four primitive,the fifth primitive—senti-mental primitive was proposed,which was integrated into HowNet manually,on the basis of this,then a new word sentimental similarity calculation method was proposed to compute word’s sentimental value. At last,combine this way with transductive learning for identif-ying word’s sentimental orientation. The performance of experiment shows that compared with SVM or traditional semantic comprehen-sion,it can get better results.

常用功能

更新日期/Last Update: 2016-04-12