摘要
通过对商品评论进行基于方面的情感分析,可以得到某件商品各个方面的优劣情况。提出利用三层CRF模型进行情感极性分类及强度分析。在CRF模型中,融合了词、词性、语气词、程度词、方面和评价词的共现等特征。在情感句识别、情感极性分类和情感强度分析上得到的F1值分别为86.3%、77.2%、70.7%,证明了:a)分层CRF模型在各个层次的任务中都能取得较好的结果;b)语气词、程度词、方面和评价词的共现特征在情感分类时的有效性。
In research about product comments, aspect-based emotion analysis can help people to find out strengths and weak- nesses of each aspect of a product. This paper proposed a sentiment analysis method based three-layered CRF model. It used all word, part of speech, modal particles, degree word, co-occurrence of aspect and evaluation term features in CRFs model. The Fl-measure of emotion sentence recognition, emotional polarity classification and emotional strength analysis were 86.3%, 77.2% and 70.7% respectively. It proves that layered CRF leads to better result in each layer than usual CRF, modal particles, degree word, co-occurrence of aspect and evaluation term are all useful features in sentiment analysis.
出处
《计算机应用研究》
CSCD
北大核心
2017年第4期986-990,共5页
Application Research of Computers
关键词
商品评论
情感分类
情感强度分析
条件随机场
product reviews
sentiment classification
sentiment strength analysis
conditional random fields
作者简介
李向前（1970-），男，江苏淮安人，副教授，硕士，主要研究方向为网络与信息系统、计算机应用技术、计算机图形图像；
李军伟（1991-），男（通信作者），硕士，主要研究方向为机器学习与认知计算（davidlee91@163．com）．
