摘 要

当今时代，网络媒介已经成为最具广泛性的一种新型传播媒介，新浪微博凭借着其便捷的特性，成为用户的首选。用户通过新浪微博获取兴趣信息并分享，同时也可以创造信息，使得新浪微博在社会网络舆情中发挥着愈发重要的作用。对新浪微博的网络舆情时空扩散的分析，为舆情监测和管制提供了一定的实际指导。

本文首先通过百度搜索引擎的数据，统计分析了各个热点话题随时间的分布情况，确定本文所研究的具有代表性的突发事件为天津爆炸事件。并获取以该突发事件为关键词、发布时间段为2015年8月12日至9月12日的微博数据。针对微博数据，在时间和地理空间两个维度上进行了统计分布，发现自事件发生开始微博的舆情热度迅速升起，并在第二天达到顶峰，然后逐渐衰减，并以事发地—天津为中心，向一线城市扩散。然后通过以隐马尔可夫模型（HMM）为理论模型的统计分词方法，将微博文本分词。运用统计学中设置权重的技术，构建TF-IDF模型，将分词的文本向量化。采用层次聚类与k-means聚类相结合的方法，对文本进行聚类，得到五个方面的主题，并识别主题的情感倾向性，得到结论是：中性的事故地点主题、偏积极性的事故主体主题、偏消极性的事故救援主题、偏积极性的事故传播主题和偏积极性的祈祷平安主题。

针对这些主题，分析微博的网络舆情，可以为政府部门监管网络舆情提供依据，在其官方微博上引导网友进行健康的网络舆情扩散，并需要调整消防队的义务兵制，将消防员职业化，既是对消防员的人身安全作保障，也是对民众负责。

关键词：文本分词；IF-IDF模型；层次聚类；k-means聚类；文本情感识别

Abstract

In the modern era, the network media has become the most extensive a new type of media. With its convenient features, micro-blog becomes the user's choice. Users access interest information and share it through micro-blog, but also create information . micro-blog plays an increasingly important role in the social networks of public opinion .Temporal diffused analysis of micro-blog's network public opinion, provides some practical guidance for monitoring and control of public opinion.

Firstly, through the Baidu search engine data, the paper counts and analyzes the various hot topics over the distribution of time to determine a representative emergencies-Tianjin bombings. And the paper gets micro-blog data of which the incident regards as a keyword from 2015 August 12 to September 12 . The paper counts micro-blog data of the incident in two dimensions of time and geographical space and founds that since the beginning heat of micro-blog's public opinion was rapidly rising, and the next day reached a peak, then gradually decreases, and spread to the first-tier cities by Tianjin as the center. Then micro-blogging text segments by Hidden Markov Models (HMM) as the theoretical model of statistical segmentation method. The word of the text is quantified by statistical techniques setting weights to construct TF-IDF model . The paper combines method of hierarchical clustering and k-means clustering and clusters text, give five topic areas and identify topics emotional propensity. The paper's conclusion is: neutral site of the accident theme, bias motivated incidents main theme, bias negativity accident rescue theme, bias motivated the spread of incident theme and bias motivated pray theme.

For these topics, micro-blog's network public opinion analysis, can provide base for government supervision network public opinion and guided users to spread health network public opinion by its official micro-blog. And government is needed to adjust the fire brigade conscripts system and make the firefighters professional. It both also ensures the safety of the firefighters and makes it responsible for people.

Key Words: Text segmentation; TF-IDF model; hierarchical clustering; k-means clustering; text emotion recognition

目录

摘 要 I

Abstract II

第1章 绪论 1

1.1引言 1

1.2本文的研究背景与意义 1

1.3国内外研究现状分析 2

1.4本文的研究技术方法和结构框架 3

1.5本文的创新点 5

第2章 研究模型的理论知识 6

2.1文本分词 6

2.1.1基于语法和规则的分词法 6

2.1.2基于词典的机械式分词法 6

2.1.3基于统计的分词法 6

2.2文本向量化 7

2.3文本聚类 8

2.3.1层次聚类 8

2.3.2 k-means聚类 9

第3章 新浪微博网络舆情实证分析 10

3.1舆情获取 10

3.1.1话题和关键词确定 10

3.1.2数据获取 11

3.2舆情分析 12

3.2.1时间与空间的分布 13

3.2.2主题确定 16

3.2.3各主题情感倾向性 22

第4章 研究结论与展望 24

4.1研究结论 24

4.2展望 24

参考文献 26

致 谢 27

第1章 绪论