摘要
[目的/意义]图书作为典型的体验型产品,缺乏与之对应的领域情感词典,限制了该领域中的用户情感分析与情感挖掘研究,文章提出一种基于超短评论的图书领域情感词典构建方法。[方法/过程]在定义超短评论概念的基础上,利用规则组合进行候选词过滤,提出一种改进点互信息的方法进行情感词极性判断,并利用用户投票方法确定词汇的情感强度。[结果/结论]在图书评论场景中,本文情感词典的准确率、召回率、词典规模均远高于通用情感词典HowNet、大连理工情感本体,新词发现能力和实际应用能力较强。
[Purpose/significance]As a typical experiential product,books lack the corresponding domain emotion dictionary,which limits the research on user sentiment analysis and sentiment mining in this field.This paper proposes a method of constructing Book domain sentiment dictionary based on extreme short comments.[Method/process]On the basis of defining the concept of ultrashort comment,the candidate words are filtered by rule combination,and an improved mutual information method is proposed to judge the polarity of emotional words,and the user voting method is used to determine the emotional intensity of words.[Result/conclusion]In the scene of book review,the accuracy rate,recall rate and dictionary scale of this emotional dictionary are much higher than those of HowNet and Dalian Institute of technology,and the ability of new word discovery and practical application is strong.
出处
《情报理论与实践》
CSSCI
北大核心
2021年第9期183-189,197,共8页
Information Studies:Theory & Application
基金
国家社会科学基金青年项目“网络知识社区用户交互内容的组织与传播研究”的成果,项目编号:18CTQ033。
关键词
情感词典
情感分析
超短评论
点互信息
图书领域
sentiment lexicon
sentiment analysis
extremeshort review
pointwise mutual information
library field