摘要
[目的/意义]对海量但杂乱的在线评论资讯进行有效的质量评测是信息领域的热点问题。文章从心理学角度挖掘评论内容蕴含的细粒度情感,"情绪"对用户行为产生的作用,为网站进行评论质量评测提供新思路,同时也为企业实施情感营销提供理论依据和参考方案。[方法/过程]研究借助心理学领域的情感模型,以长篇幅"书评"为研究对象,运用自然语言处理技术从评论内容中提炼出"乐、哀、怒、惧、惊、恶、好"7种细粒度情绪特征及主观性特征,采用基于随机森林的分类和特征优选算法,对情绪特征对评论感知效用的预测力进行实证检验,深入分析细粒度情感对评论感知效用的影响。[结果/结论]结果表明,评论内容中蕴含的细粒度情感与评论感知效用关系密切,情绪特征的引入能够显著提升感知效用预测模型的预测力。情感语义纯粹的"乐、怒、恶"作用更显著,"乐"带给读者的愉悦易使评论获得有用性评价,"怒、恶"的作用则相反;情感语义复杂的"好、惊、惧、哀"与评论感知效用的关系相对较弱;但情绪的影响力会随评论对象(图书)的不同而不同。针对评价对象构建基于情感特征的评论有用性预测模型更有实际意义。
[Purpose/significance]Review quality evaluation is one of hot issues in information area.This paper investigates the impact of emotions on review helpfulness from psychology perspectives,providing a new strategy for helpful review identification,as well as theory references for enterprise marketing.[Method/process]Emotion influence on review helpfulness is explored on the basis of an emotion theory in psychology.For target objects,long reviews,sentimental features in the review content is extracted firstly,including seven specific emotions(eg.joy,sadness,anger,fear,surprise,disgust,goodness)and two subjectivity indexes.Five Random-forest models based on different sentimental feature combination are then generated to identify helpful reviews.By analysis the five classifiers’performance,impacts of emotions on review helpfulness are able to be estimated.[Result/conclusion]The results indicate emotions are significantly related with review perceived helpfulness.Straight forward emotions,such as"joy""anger"and"disgust",are more significant than complicated emotions(e.g.goodness,surprise,fear and sadness)."Joy"promote perceived helpfulness of reviews,where as impacts of"anger"and"disgust"are negatively.Additionally,the results show that emotion influence is associated with evaluated objects(e.g.books)as well,suggesting it’s more significant to generate a sentiment-oriented helpful reviews predictor for specific objects being reviewed.
出处
《情报理论与实践》
CSSCI
北大核心
2019年第9期117-122,153,共7页
Information Studies:Theory & Application
基金
国家社会科学基金项目“面向用户感知效用的在线评论的质量与控制研究”的成果之一,项目编号:15BTQ067
关键词
细粒度情感分析
评论感知效用
情感模型
情感词典
在线评论
fine-grained sentiment analysis
review perceptive helpfulness
emotion model
sentiment lexicon
online review