一种面向旅游评论的情感特征识别方法

A Recognition Method of Sentiment Features for Tour Reviews

情感分析的一大难点是如何获取主题相关的情感特征信息。首先给出了"有效"情感特征的定义,然后提出了一种基于语义角色标注的有效情感特征抽取方法。该方法先依据评论库的主题元数据得到候选主题特征项,并标注主题句,然后执行主题句的语义角色标注,基于情感特征所在的角色类型判断该特征语义是否指向主题项。该方法的特点在于过滤与指定主题无关的噪声特征。实验面向旅游景区游客评论在不同规模的标注集环境下对比了三种特征抽取方法,即基于词袋的方法、基于主题的方法和文中基于有效情感特征的方法,结果显示文中方法对于短文本的情感分类较词袋方法有3%的性能提升,而对于长文本的情感分类其性能较前面两种方法优势突显,总体达到了84.81%的准确率。

One of the challenges faced by sentiment analysis is how to obtain the information of sentiment characteristics related to the subject.We give the definition of effectiveness of sentiment features firstly.Then we propose an effective sentiment feature extraction method based on semantic role annotation.This method first gets candidate topic feature items based on the topic metadata of the comment library and annotates topic sentences.Then,semantic role annotation of topic sentences is performed,and the semantic of the feature is judged to be directed to the topic item based on the role type of the sentiment feature.This method is characterized by filtering noise features that are irrelevant to the specified subject.In experiment on scenic tourist comments in different sizes of tagging set environment,we compare the three methods of feature extraction respectively based on word bag,based on topic and based on effective features in this paper.It shows that the proposed method has 3%performance improvements for short text classification of sentiment than the word bag method,and for a long text sentiment classification,its performance is remarkably highlighted with 84.81%of overall accuracy.