面向中文微博文本的情感极性判别方法研究

发布时间：2018-06-01 06:34

本文选题：微博 + 情感分析　；参考：《中国民航大学》2017年硕士论文

【摘要】：微博是一个分享实时短信息的社交媒体,为网民提供了一个沟通思想,交流看法的平台。用户既可以作为一个观众来浏览感兴趣的话题信息,也可以作为一个内容发布者提供内容供其他人浏览。这种新的社交形式已经被大众广泛接受,在近十年中取得了用户数量和信息量的爆发式增长。微博文本涉及到的话题广泛,为情感分析研究提供了丰富的语料素材。情感极性判别是情感分析的一种具体任务,通过对带有情感色彩的文本进行处理、分析,从而判断它的情感是属于积极的还是消极的。对微博进行情感极性判别的目的在于识别出微博中的主观信息,从而挖掘出用户对热点话题、新闻事件和产品服务等评价对象的观点和态度,从而实现舆情监控、市场营销等目的。本文选取包含民航舆情关键词的微博进行情感极性判别研究,分别给出了基于情感词和语义规则、基于Adaboost和分类器加权投票的两种方法。第一种方法结合现有情感词典和语义相似度计算来提取情感词,避免了情感词典中词语不全而导致情感词遗漏的情况,然后利用句内规则和句间规则计算出微博文本情感得分,与微博表情符号得分做加权求和,得到最终情感极性分值,从而判断微博的情感极性。第二种方法基于机器学习判别情感极性,利用Adaboost集成学习对效果较差的单一分类器进行提升后,将三种分类器进行加权投票组合,得到最终分类器对测试微博文本进行分类。本文所使用的微博语料是利用微博爬虫抓取的,对其情感极性进行人工标注后用于实验,实验结果表明:两种方法在情感极性分类的准确度上均取得了较好的效果,第一种方法不依赖于领域知识,对微博内容的普适性较强;第二种方法使弱分类器的性能得到提高,结合了不同分类器的优势,克服了单一分类器的缺陷。
[Abstract]:Weibo is a social media that shares real-time short messages and provides a platform for Internet users to communicate ideas and ideas. Users can view topics of interest either as an audience member or as a content publisher for others to browse. This new form of social networking has been widely accepted, in the past decade has achieved explosive growth in the number of users and information. Weibo texts cover a wide range of topics and provide rich corpus material for affective analysis. Emotion polarity discrimination is a kind of concrete task of emotion analysis. By processing and analyzing the text with emotion color, we can judge whether its emotion is positive or negative. The purpose of judging the emotional polarity of Weibo is to identify the subjective information in the Weibo, so as to excavate the opinions and attitudes of the users on the hot topics, news events and products and services, so as to realize the monitoring of public opinion. Marketing, etc. In this paper, we select the Weibo which contains the key words of civil aviation public opinion to study the emotional polarity, and give two methods based on emotional words and semantic rules, Adaboost and classifier weighted voting. The first method combines the existing emotion dictionary and semantic similarity calculation to extract the emotion words, which avoids the omission of the words in the emotion dictionary, and then calculates the emotional score of the Weibo text by using intra-sentence rules and inter-sentence rules. The final emotional polarity score was obtained by weighted summation with the Weibo emoji score, and then the emotional polarity of Weibo was judged. The second method is based on machine learning to distinguish emotional polarity. After using Adaboost ensemble learning to improve a single classifier with poor effect, the three classifiers are weighted to vote together, and the final classifier is obtained to classify the test Weibo text. The Weibo corpus used in this paper is captured by Weibo crawler, and the affective polarity is labeled manually for experiment. The experimental results show that the two methods have achieved good results in the classification accuracy of affective polarity. The first method does not rely on domain knowledge and has a strong universality for Weibo content. The second method improves the performance of weak classifier and combines the advantages of different classifiers to overcome the shortcomings of single classifier.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1;TP393.092

【相似文献】