导　　师： 钱国明

学科专业： L01

授予学位： 硕士

作　　者： ;

机构地区： 哈尔滨工业大学

摘　　要： 突发事件具有不确定性、影响范围广、后果严重等特点，一旦发生就会引起广泛关注。随着论坛、微博等网上交流平台的广泛使用，突发事件的网络舆情会以极快的速度形成，并在短时间内迅速扩散。通过对网络舆情的分析，不仅可以了解公众的态度与看法，而且可以从中发现突发事件应急处理工作中的不足，为工作的及时改进以及将来制定决策提供依据。 通常来说，突发事件网络舆情的评价对象不仅仅是事件的本身，还包括政府部门、新闻媒体等方面。针对不同的评价对象，公众在表达情感时使用的词语可能存在差别，而且，评价对象不同的网络舆情在演化的过程可能也会呈现出不同的特点，因此，应当根据不同的评价对象分别进行情感倾向及舆情演化分析。 为了对突发事件的网络舆情进行情感倾向及演化分析，本文首先从天涯论坛的天涯杂谈版块中获得了所需的数据，并利用文本聚类技术将所有评论划分为政府部门类、新闻媒体类、事故原因类和其他类，然后基于由《知网》提供的情感词构建的情感词典，对每一类的评论进行情感分析。将不同的特征提取算法与分类算法组合进行多组情感分析实验，结果表明信息增益特征提取算法和支持向量机分类算法的分类效果最好。基于情感分析的结果，本文针对不同类别的评论分别对网络舆情进行演化分析。将热点话题结合到演化分析的过程中发现：热点话题有助于解释情感变化的原因。本文在演化分析的基础上探究积极评论人对网络舆情的情感导向作用，结果表明积极评论人在政府部门类舆情中对消极情感具有情感导向作用。 Since emergency owns the characteristics of uncertainty, wide impact and seriousconsequence, emergency becomes the focus of public attention. With the widespread ofonline communication platform, like forums and micro-blogs, network public opinionon emergency forms soon and spreads fast. Based on the analysis on public opinion, wecan understand public attitudes and figure out the insufficient of emergency responseand handling work, which turns to be the basis of work improvement anddecision-making in the future. Generally, the comment objects of emergency network public opinion include notonly the event but also the government, media, etc. For different comment objects, thesentiment words used by the public may differ and the evolution of network publicopinion may have different characteristics. Hence, the sentiment analyses and theevolution analyses of public opinion need to be performed respectively by differentobjects. As to analyze the sentiment orientation and its evolution of public opinion onemergency, this paper, in the first place, collects data from Tianya By-talk of TianyaForum and clusters the comments into four categories by using text clustering. Thecategories are government department, media, the cause and others. Then, this paperanalyzes the sentiment orientation of online reviews based on a sentiment lexicon whichis built with sentiment words provided by HowNet. From the experiment results withdifferent feature extraction methods and text classification methods, it indicates that theprecision of information gain and SVM is higher than others. Based on sentimentanalysis, this paper analyzes the evolutions of network public opinion in differentcategories. The evolution analysis aligning with hot topics shows that the topics havecontributed to explain the reason of emotional changes. By analysis on the relationshipbetween positive commentators' opinion and public opinion, I find that positivecommentators have an emotional guidance effect on negative emotion of public opinionin the reviews of government department class.

关 键 词： 突发事件 网络舆情 情感词典 情感分析 舆情演化

分 类 号： [TP391.6]

领　　域： [自动化与计算机技术] [自动化与计算机技术]