面向民航评论的情感分析方法及应用研究

发布时间：2018-03-30 09:23

本文选题：情感分析　切入点：文本分类　出处：《中国民航大学》2017年硕士论文

【摘要】：随着微博等自媒体平台的迅猛发展,互联网传播信息的速度已经超越传统媒体成为最具影响力的信息传播途径,因此社会热点事件的传播速度越来越快,影响范围也越来越广。民航业在其快速发展过程中不免暴露出各种问题,这些问题同样经由信息手段快速传播,在社会上极易引起强烈反响。如何有效地从中抽取出我们想要的信息并进行数据分析,进而对这些社会热点事件情感倾向做出分类并识别这些热点事件中的评论对象,已经成为舆情传播的重要研究内容之一。本文主要的研究对象为微博评论文本,我们从海量评论文本中抽取与民航相关的数据,将其情感倾向进行人工标注后,研究基于支持向量机的情感分类方法。对于分类器而言,特征提取方案是否优良,是决定训练结果的重要指标之一,因此本文考虑了两种特征提取方案,分别为词频特征和词语情感属性特征,均得到了较好的分类结果,正确率达到百分之九十以上。针对负面评论,确定其中的指责对象具有重要的研究意义。本文研究了评论对象分类问题,由于传统的信息增益方法只考虑类别对特征的影响,而大量短微博中的特征很可能会被忽略,影响分类效果,因此本文在信息增益方法的基础上加入文本长度对特征的影响,在新浪微博上的实验表明,新方法的分类准确度较传统的信息增益方法有所提高。本文还研究了基于Labeled LDA模型的针对不同评论对象时所用的词语,研究了基于吉布斯采样结果的评论用语之间的关系,并构建了针对不同评论对象的用语关系网络,形象地展示了这些评价词语之间的关联关系。
[Abstract]:With the rapid development of Weibo and other self-media platforms, the speed of information dissemination on the Internet has surpassed the traditional media and become the most influential way of information dissemination. Therefore, the speed of transmission of social hot events is getting faster and faster. The civil aviation industry, in its rapid development, inevitably exposed problems that also spread rapidly through information, How to effectively extract the information we want and analyze the data, and then classify the emotional tendency of these social hot events and identify the comment objects in these hot events. It has become one of the important research contents of public opinion communication. The main research object of this paper is Weibo comment text. We extract the data related to civil aviation from the massive comment text, and label its emotional tendency manually. This paper studies the emotion classification method based on support vector machine. For classifier, whether the feature extraction scheme is good or not is one of the important indexes to determine the training result. Therefore, two feature extraction schemes are considered in this paper. For word frequency feature and word affective attribute feature, good classification results are obtained, and the correct rate is more than 90%. In view of negative comments, It is of great significance to determine the target of accusation. In this paper, we study the classification of comment objects. Because the traditional information gain method only considers the influence of categories on features, a large number of features in short Weibo are likely to be ignored. The effect of classification is affected, so this paper adds the influence of text length to feature based on the information gain method, and the experiment on Sina Weibo shows that, The classification accuracy of the new method is higher than that of the traditional information gain method. This paper also studies the words used for different comment objects based on the Labeled LDA model, and studies the relationship between the comment terms based on Gibbs sampling results. A network of lexical relationships for different comment objects is constructed, which vividly shows the correlation between these appraised words.
【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

【参考文献】