基于AF模型的跨领域情感词典构建方法研究

发布时间：2018-05-19 05:22

本文选题：情感倾向 + 跨领域　；参考：《华中科技大学》2016年硕士论文

【摘要】：随着web2.0的快速发展,用户变成了互联网内容贡献的主力,大量用户产生的主观性文本充斥了人们的视野。基于这些主观性文本的情感分析成为了近些年的研究热点。情感词典的构建是情感分析的基础工作,然而构建情感词典时存在不平衡语料问题,所以,针对这个问题的跨领域情感词典构建研究也正在获得越来越多人的关注。目前的跨领域情感词典构建方法大都需要很多的人工标注信息,这些标注信息需要投入大量的人力和时间才能得到,而且还不够可靠,针对这个问题,提出一种新的基于AF模型的跨领域情感词典构建方法,该方法仅需要源领域中有词语级的标注信息,就可以利用词语之间的上下文相似度来判断目标领域中词语的倾向性,从而构建情感词典。此构建方法主要分为三个内容,第一个内容是语料数据的预处理,主要对语料中的句子进行断句,分词等处理;第二个内容是源领域和目标领域AF模型的建立,将两个领域语料的预处理结果建立成AF模型;第三个内容是目标领域情感词倾向性的判定,此为所提出的构建方法的核心内容,其中基于AF模型提出了两个新的统计量——领域差异度和相对亲和度。领域差异度用来区分领域相关情感词和领域无关情感词,相对亲和度用于计算两个领域间的情感词的上下文相似度。在这两个统计量的基础上,结合建立的模型,给出一个倾向性判定算法。通过倾向性的判定结果,将判断出的正向情感词和负向情感词结合在一起就构成了目标领域的领域情感词典。最后,在COAE2011提供的评测数据基础上,将所提出的词典构建方法与SO_PMI方法、以及融合上下文的构建方法进行对比分析,验证本方法的有效性。除此之外,通过实验不同参数对所提出构建方法的结果影响,来分析此跨领域词典构建方法的特性。
[Abstract]:With the rapid development of web2.0, users have become the main contribution of Internet content, and a large number of subjective texts produced by users are flooded with people's vision. Emotional analysis based on these subjective texts has become a hot topic in recent years. The construction of emotion dictionary is the basic work of emotion analysis. However, there is an unbalanced corpus problem in the construction of emotion dictionary. Therefore, more and more people are paying attention to the cross-domain emotion dictionary construction. At present, most of the cross-domain affective dictionary construction methods need a lot of manual tagging information, which needs a lot of manpower and time to get, and is not reliable enough, in order to solve this problem, A new cross-domain affective dictionary construction method based on AF model is proposed. This method only requires tagging information at the word level in the source domain, and can use the contextual similarity between words to judge the tendency of the words in the target domain. Thus construct the emotion dictionary. The method is mainly divided into three parts: the first is the preprocessing of corpus data, the second is the construction of AF model in the source and target fields, which mainly deals with sentence breakage and word segmentation in the corpus. The preprocessing results of the two domain corpus are established into AF model. The third is the judgment of the tendency of emotional words in the target domain, which is the core content of the proposed construction method. Based on AF model, two new statistics, domain difference degree and relative affinity degree, are proposed. Domain difference is used to distinguish domain-related affective words from domain-independent affective words, and relative affinity is used to calculate the contextual similarity of affective words between two domains. On the basis of these two statistics and combined with the established model, a decision algorithm of tendency is given. Through the judgment result of tendency, combining the positive affective words and negative affective words to form the domain emotion dictionary of the target domain. Finally, on the basis of the evaluation data provided by COAE2011, the method of dictionary construction is compared with the method of SO_PMI, and the method of constructing fusion context is compared to verify the validity of this method. In addition, the characteristics of this cross-domain dictionary construction method are analyzed by the influence of different parameters on the results of the proposed method.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.1

【参考文献】