中文微博情感倾向性分析特征工程

发布时间：2018-02-28 00:26

本文关键词： 情感倾向性分析中文微博特征工程　出处：《山西大学学报(自然科学版)》2014年04期 　论文类型：期刊论文

【摘要】：情感倾向性分析是情感分析的重要组成部分,是一种按照情感倾向对文本进行分类的任务。微博与传统的评论文本相比更加口语化与符号化,因此对微博进行情感倾向性分析是一个非常有挑战性的任务。基于机器学习的方法是情感倾向性分析最经典的算法,核心是要进行特征的分析和选择,例如词袋特征等。然而,由于中文语言的独特性,前人很多有效的特征都是语言相关的,将其直接用于中文微博效果不佳。在中文微博语料上,还没有学者进行细致的特征工程建设。基于此,文章综合国内外诸多特征,并考虑到中文的独特性,对中文微博的褒贬中倾向性判别特征工程的词、词组、数值和句法特征分别进行了研究,并提出了基于词典规则的情感评分的新特征。最后经过大量实验与分析,得出了可靠的特征组合。实验结果表明,此方法能够明显提高情感倾向性分析的结果。
[Abstract]:Affective orientation analysis is an important part of affective analysis and a task of classifying texts according to emotional tendency. Weibo is more colloquial and symbolic than traditional comment texts. Therefore, it is a very challenging task to analyze Weibo's affective orientation. The method based on machine learning is the most classical algorithm of affective orientation analysis, the core of which is to analyze and select features, such as word bag features, etc. However, Because of the uniqueness of the Chinese language, many of the former effective features are language-related, and their direct application to the Chinese language Weibo is not very effective. No scholar has carried out a detailed feature engineering construction in the Chinese Weibo corpus. Based on this, This paper synthesizes many characteristics at home and abroad, and taking into account the uniqueness of Chinese, studies the words, phrases, numerical values and syntactic features of Chinese Weibo's favoritism distinguishing feature engineering. A new feature of emotion score based on dictionary rules is proposed. After a lot of experiments and analysis, a reliable feature combination is obtained. The experimental results show that this method can obviously improve the results of emotional orientation analysis.
【作者单位】：哈尔滨工业大学计算机科学与技术学院信息检索研究中心;哈尔滨工业大学机电学院媒体系;
【基金】：国家自然科学基金重点项目(61133012);国家自然科学青年基金项目(61300113);国家自然科学基金面上项目(61273321)
【分类号】：TP391.1;TP393.092

【参考文献】