基于微博平台的中文情感分析技术的研究

发布时间：2018-05-09 21:48

本文选题：微博 + 情感分析　；参考：《沈阳工业大学》2017年硕士论文

【摘要】：随着互联网的快速发展与WEB2.0时代的到来,微博、微信等社交网络平台在人们生活中占有越来越重要的位置。互联网上的博文数量呈爆炸式增长、海量的博文信息中包含着用户的情感、观点,可为政府机构舆情分析、企业市场行为策略制定、消费者购买行为分析等方面提供决策数据支持。对微博文本进行情感分析,发掘其中潜藏的巨大价值具有重要的商业及社会意义。基于微博文本的情感分析的主要工作是判别微博文本的情感倾向性。本文首先构建了微博情感分析用情感词典,基础情感词典部分采用对现有的中文情感词典NTUSD和HowNet两者的综合,随后通过整理微博语料中的表情符号、新词与网络流行词语等使用一种PMI与word2vec综合的方法对情感词典进行了扩充,得到情感词典。然后从网络上获取微博文本语料数据与第六届中文倾向性分析评测(COAE2014)评测语料数据并对其进行分词、停用词处理等预处理,之后采用人工标注与现有已标注情感倾向语料文本结合的方式构建初始训练集。本文阐述了一种综合文本情感分类方法,该方法克服了基于情感词典的分类方法过度依赖情感词典,对于未登录词处理能力欠佳的问题,以及基于机器学习的分类方法构建特征向量时丢失文本中的程度副词、句型句式关系等情感分析相关要素的缺点。本文的分类方法在构建训练用特征向量空间的过程中将传统支持向量机分类方法中丢弃的信息予以保留,融入到特征向量之中,因其在分类阶段使用基于机器学习的算法,在一定程度上保留了对情感词典中未登录词的处理能力最后,通过微博文本语料对本文的分类方法进行了验证,结果表明较传统的情感分析方法表现出更高的准确率。
[Abstract]:With the rapid development of the Internet and the arrival of the WEB2.0 era, Weibo, WeChat and other social networking platforms play an increasingly important role in people's lives. The number of blog posts on the Internet is increasing explosively. The massive amount of blog information contains the user's emotion and viewpoint, which can be used to analyze the public opinion of government agencies and make the strategies of enterprise market behavior. Consumer purchase behavior analysis and other aspects to provide decision data support. It is of great commercial and social significance to analyze Weibo's text and explore the potential great value. The main work of emotion analysis based on Weibo's text is to judge the emotional tendency of Weibo's text. This paper first constructs an emotional dictionary for Weibo's affective analysis. In the part of the basic emotional Dictionary, the author combines the existing Chinese emotion Dictionary (NTUSD) and the Chinese emotion Dictionary (HowNet), and then arranges the emoticons in Weibo corpus. The new words and popular words are extended to the emotion dictionary by using a comprehensive method of PMI and word2vec, and the emotion dictionary is obtained. Then the text data of Weibo and COAE2014 are obtained from the network and processed by word segmentation, word discontinuation, etc. Then the initial training set is constructed by the combination of manual tagging and existing tagged affective tendency corpus. This paper presents a comprehensive text affective classification method, which overcomes the problem of over-reliance on affective dictionary and poor processing ability of unrecorded words based on affective dictionary. And the shortcomings of affective analysis such as degree adverb and sentence pattern relation in the text are lost when the classification method based on machine learning is used to construct the feature vector. In the course of constructing the feature vector space for training, the classification method in this paper preserves the information discarded in the traditional SVM classification method and integrates it into the feature vector, because it uses the machine learning algorithm in the classification stage. To a certain extent, the processing ability of unrecorded words in emotion dictionary is retained. Finally, the classification method of this paper is verified by Weibo text corpus, and the result shows that the accuracy of this method is higher than that of traditional emotion analysis method.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1

【参考文献】