机构地区: 中国科学院计算技术研究所
出 处: 《五邑大学学报(自然科学版)》 2012年第4期66-71,共6页
摘 要: 将句法平面词的词性特征、依存关系、依存关系中的词性特征、邻接依存关系、邻接依存关系中的词性特征与倾向性词汇和倾向性搭配作为支持向量机(SVM)分类器的特征集,以句子为单位对多个领域的文本进行倾向性判断.通过交叉验证的方式,估计出分类器的精度为95.6%,据此提出句子倾向性分析可不以句子倾向性判断为前提. The objective sentences of multi-domain from views is distinguished by using part of speech, dependency relationship, the part of speech combinations of the two words under one dependency, two adjacent dependencies, the part of speech combinations of the three words under two adjacent dependencies, sentiment words and sentiment collocations as features of SVM classifier. The precision is about 95.6% with 10-fold cross-validation. It is assumed that the sentence tendency judgement is not the premise of the document sentiment analysis.
领 域: [自动化与计算机技术] [自动化与计算机技术]