基于图挖掘的推特事件关联性分析方法研究

发布时间：2018-06-26 11:26

本文选题：推特 + 社团检测　；参考：《电子科技大学》2017年硕士论文

【摘要】：随着互联网的发展,社交媒体在人们生活中的应用越来越多样化。而推特(Twitter)作为社交媒体中的佼佼者,已经成为近年来最流行的社交媒体应用之一。而社交媒体在政治事件中的影响也在与日俱增。2012美国总统奥巴马连任、英国脱欧公投等一系列政治事件的背景中都出现了推特的身影。推特在事件传播、反映民众的政治倾向上有逐步取代传统民意调查的趋势。目前在社交媒体上的政治倾向研究主要针对文本中的特定信息来进行分析,例如推文中的标签(hashtag)、提到(@)等。由于社交媒体数据不具有正式性,所以政治倾向分析的结果不够精确。同时对于社交媒体中的大选选情预测并没有特别完善的流程和方案。所以,本文通过图挖掘的方法,对社交媒体中的政治类事件进行分析研究。针对2016美国总统大选选情预测的问题,提出并设计了大选选情预测模型。本文的主要工作和创新点概括如下:(1)在推特数据上进行政治倾向情感分析和大选相关事件检测。在情感分析上,针对推特信息简短、非正式、缺乏补充信息的特点,本文采用了基于字典的情感分析方法,对推文的政治倾向性进行判断。同时对于情感分析中的反语鉴别难点,通过推文的表情符以及用户的历史推文来提升情感分析的结果。在事件检测中,针对美国总统大选的背景,本文采用了多次聚类、同义词拓展,关键词权重提升等方法,来对碎片事件进行整合。该方法提升了大选相关事件检测的性能。(2)利用图挖掘的方法来对推特数据进行研究分析。由于推特数据源上有多种多样的信息,例如用户,推文,图片,视频等等。而用户的点赞,转发和评论行为,往往也会表露出用户的政治倾向。所以,本文采用复杂网络分析方法,将不同类型的社交媒体数据投影到复杂网络中,再根据实际需求,对社交媒体复杂网络进行分析。在大选预测结果中发现支持总统候选人的用户社团。这种分析方法不仅仅适用于大选选情预测,还可以应用于社交媒体的舆论导向分析、用户影响力等方面。本文使用了真实社交媒体数据,分别对情感分析方法,事件检测模型和大选预测模型进行实验。实验结果显示,本文提出的大选预测模型可以对大选选情进行正确预测。
[Abstract]:With the development of Internet, the application of social media in people's life is more and more diversified. Twitter, a leader in social media, has become one of the most popular social media applications in recent years. Social media influence in political events is also growing. 2012 U.S. President Barack Obama re-elected, the British Brexit referendum and other political events in the background of Twitter. Twitter's spread of events reflects a gradual displacing of traditional opinion polls from popular political tendencies. The current research on political orientation on social media is mainly focused on the analysis of specific information in the text, such as the tag (hashtag), mentioned in Twitter (@), and so on. Because social media data are not formal, the analysis of political tendencies is not accurate enough. At the same time, there is no particularly sound process and program for social media election predictions. Therefore, this paper analyzes the political events in social media by the method of graph mining. Aiming at the problem of presidential election prediction in 2016, this paper puts forward and designs a model of presidential election prediction. The main work and innovations of this paper are summarized as follows: (1) the analysis of political tendency and emotion and the detection of election-related events on Twitter data. In terms of affective analysis, in view of the features of short, informal and lack of supplementary information, this paper adopts a dictionary-based emotional analysis method to judge the political tendency of Twitter. At the same time, for the difficulty of identifying irony in affective analysis, the result of emotional analysis is improved by the emoji of tweets and the historical tweets of users. In the event detection, in view of the background of the American presidential election, this paper uses several methods, such as clustering, synonym extension, keyword weight enhancement, to integrate the debris events. This method improves the performance of election related event detection. (2) the method of graph mining is used to study and analyze Twitter data. Because Twitter data sources have a variety of information, such as users, tweets, pictures, videos and so on. Users' likes, retweets and comments often show their political inclination. Therefore, this paper uses the method of complex network analysis, projects different types of social media data to complex network, and then analyzes the complex social media network according to the actual demand. In the election forecast results found in support of the presidential candidate user community. This analysis method not only applies to election prediction, but also can be applied to social media opinion oriented analysis, user influence and so on. In this paper, real social media data are used to test affective analysis, event detection model and general election prediction model. The experimental results show that the proposed general election prediction model can correctly predict the election results.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O157.5;TP391.1

【相似文献】