基于HNC语境框架和情感词典的文本情感倾向分析

提出了一种基于情感词典和概念层次网络(hierarchical network concepts, HNC)语境框架的文本情感倾向性分析方法,将文本的情感倾向分析分为两个阶段:特征词、语句和句群判定阶段;基于HNC语境框架的句与句群情感分析阶段。首先以HowNet情感词典和自建的形容词配价词典(valency dictionary of English adjective, VDEA)作为基础词典资源进行文本特征词匹配,在此基础上基于HNC语境框架进行文本的情感倾向性判定,融合情感词典资源与HNC语境框架的独特优势,从特征词语情感分析入手,以包含特征词的语句及句群为情感分析重点,进而确定文本的情感倾向性,体现了HNC “有所为有所不为” 的思想。为验证方法的有效性,文本分别对政治、经济、体育与影视评论等领域文本进行测试,从实验结果可以看出商品评论以及影评类的文本情感识别率相对较高,而政治与体育类识别率低,但基本达到了预期实验效果,从而验证了本方法的可行性。

Text sentimental orientation analysis based on HNC contextual framework and sentimental dictionaries

Based on the current sentimental dictionaries and HNC contextual framework, a method of text sentimental orientation analysis was put for ward. The sentimental analysis process covers two phases: feature words matching and feature sentence(or sentence group)finding; feature sentence or sentence group sentimental analysis based on HNC contextual framework. In the first phase, sentimental dictionary of HowNet and Valency Dictionary of English Adjective(VDEA)are applied while the feature sentence or sentence group are analyzed in the second phase. The method, through exact matching of feature words, makes the posterior processing work more effective energy-focused because only those sentences or sentence groups containing subjective sentiment can be analyzed and processed. This thought also illustrates one of the spirits of HNC: doing certain things and refraining from doing other things. The paper takes texts concerning politics, economy,sports and films’ comments as experimental data and experiment result shows that sentimental orientation recognition rate of texts concerning goods and film comments are higher than that politics and sports. The expected experimental results of the paper is reached, which tested the feasibility of the method.