联系我们

论文查询结果	返回搜索

论文编号：	9339
作者编号：	1120130831
上传时间：	2017/6/20 13:50:20
中文题目：	以事件为中心的网络舆情演化研究
英文题目：	Study on the Evolution of Event-Centered Online Public Opinion
指导老师：	王芳 教授
中文关键字：	网络舆情演化，议程设置，事件情节，话题追踪，情感分析
英文关键字：	Online Public Opinion Evolution，Agenda Setting，Event Storyline，Topic Tracking，Sentiment Analysis
中文摘要：	随着社交媒体的发展，网民参与讨论和扩散舆情事件的热情愈加高涨，网络舆论对现实社会的影响也不断增强。一方面，在网络群体效应的作用下，以事件为中心的网络舆情很容易出现意见分化和情感聚集的群体极化现象，从而导致严重的现实后果。另一方面，公众在社交媒体中所表达的意见和建议能够推动事件的处理，帮助政府准确回应网络参与者的疑惑和不满。针对事件的媒体报道和大众在社交网站中对事件的关注共同推动着舆情的发展，然而，传统的网络舆情分析往往只关注舆情在社交媒体中的演化，却忽略了引发舆情的事件本身在媒体报道中的发展，而且面对愈加庞大的网络文本数据，基于样本的内容分析也限制了我们对舆情演化的解读。基于以上背景，本文首先以议程设置理论为指导，以文本挖掘方法为基础，建立了以事件为中心的网络舆情分析框架。然后，深入到网络舆情的事件层、话题层以及观点层，通过实证案例分别探索了不同层次各自的演化过程。最后，从议题和特征两个层面研究了以事件层为代表的媒体议程与话题层和观点层为代表的公共议程之间的互动关系。 本文的主要研究工作及结论包括：第一，改进了从新闻报道集合中提取事件情节与互动关系的方法，利用时间连贯性、主体亲和性和子话题一致性来计算事件关联度，并通过案例事件数据建立了事件关联网络，分析了事件情节的演化模式。研究发现，事件情节互动结构提取的方法能够清晰地展现完整的事件脉络，帮助网络参与者和政策制定者更好地理解和把握事件报道的发展。同时，事件情节的演化过程呈现伞状结构，情节之间会彼此抢夺媒体的关注焦点地位。最后，新闻媒体会通过重复报道来加强公众对于某些子事件的认知，但在新闻媒体平台之间并不存在议程争夺的现象。第二，引入动态主题模型对事件相关的微博文本进行主题建模，抽取了舆情演化过程中主要的子话题，将舆情分析深入到了词语表达的细粒度层次。研究发现，从子话题的内容上来看，大多数微博关注的是对事件进展的讨论，而且充满了人文关怀的色彩，从关键词的变化来看，政府对事件的积极回应态度能够极大的缓解舆情的恶化，而尝试延迟报道或消极面对则不利于舆情的引导。第三，提出词性标注和句法分析抽取微博参与者对事件实体观点的方法，并利用词云进行了微博观点画像。提出了利用Word2Vec方法扩展情感词典的方法，并从情感极性和类别两个方面分析了大众情感的演化。研究发现，“疑惑”、“关注”和“诋毁”往往是公共安全事件舆情中常常存在并占据主导地位的情感。而且“疑惑”会拉长舆情的周期，扮演着引发争论的带动者角色。第四，运用格兰杰因果分析检验了媒体议程与公共议程之间的互动关系，并探索了新闻报道和微博所表达的情感类型之间的相关关系。结果发现，媒体议程影响了公共议程的议题，而公共议程则影响着媒体议程的议题特征。同时，在“庆安枪击事件”的情境下，新闻报道中所使用的情感词与微博中的情感类型具有显著的正相关关系。 本研究的主要改进和创新体现在以下三方面：第一，将事件相关的新闻报道数据和社交媒体数据整合到网络舆情演化分析中，在议程设置理论的指导下，从传统舆情分析的数量统计层面延伸到了事件层、话题层、观点层的内容层面，从议题议程设置的层面深入到了特征议程设置的层面，完善了舆情演化的分析框架。第二， 改进了动态追踪事件情节的方法，并利用社区探测方法自动提取事件关联网络中的焦点子事件，从而建立事件演化的情节互动结构。引入动态主题模型提取微博子话题，从话题热度和子话题的词语变化两个层面探索了话题的演变。提出了微博观点画像的方法并引入Word2Vec对情感词典进行扩充，弥补了网络舆情演化分析的不足。第三，探索了在不同事件情境下，网络舆情不同层次之间以及内部的互动关系，为未来分析舆情演化、集群行为和情感传播提供了理论基础。 图27幅，表27个，参考文献210篇。
英文摘要：	With the development of social media, the growing enthusiasm of Internet users’ participation in the discussion and spread of public events has embolden the impact of public opinion on real society. On the one hand, under the influence of the network clustering effect, the event-centered online public opinion is prone to the polarization of opinion and emotional aggregation, which leads to serious realistic consequences. On the other hand, the opinions and suggestions expressed by the public in the social media can facilitate the handling of the event and help the government to respond accurately to the users’ doubts and dissatisfaction. Media coverage and public concern in social media have jointly promoted the dynamic change of public opinion. However, the traditional online public opinion analysis is often concerned only with the evolution of public opinion in social media, while ignoring the development of event itself. Moreover, with the increasing amount of internet text data, content analysis based on sample extraction limits our interpretation of public opinion evolution. In the above background, based on agenda setting theory and text mining, this paper first established an event-centered network public opinion analysis framework. Then, we extend to event layer, topic layer and view layer and through the empirical case respectively explored the different levels of their own evolution process. Finally, the interaction between the media agenda represented by the event layer and the public agenda represented by the topic layer and the view layer is studied from the two aspects of the issue and the attribute. The main research work and conclusions are as follows: Firstly, the method of extracting the event storylines and interaction from the news report collection is improved. The event correlation degree is calculated by using time coherence, subject affinity and sub-topic consistency, and event-related network is established through case event data, and the evolution pattern of event plot is analyzed. The study finds that the extraction method of event storyline interaction structure can clearly show the complete event context and help users and policy makers to better understand and grasp the development of event reports. At the same time, the news media will repeat the report to strengthen the public awareness of certain sub-events, but there is agenda competition between news platforms. Secondly, the dynamic topic model is used to analyze the microblogging text related to the event, and the main subtopics in the process of public opinion evolution are extracted. From the content of the subtopics, most of the discussions focus on the event progress and full of humanistic. Viewing from the keyword changes, the government's positive attitude response to the event can greatly ease the deterioration of public opinion, and try to delay the report or negative reaction is not conducive to the guidance of public opinion. Thirdly, we propose a method using lexical annotation and syntactic analysis to extract the microblog opinion orientation of the event entities, and use the word cloud to show the microblogging view portraits. This paper puts forward the method of extending the emotional dictionary by using the Word2Vec, and analyzes the evolution of the public emotion with the emotional polarity and sentiment categories. The results show that "puzzles", "concerns" and "slander" are often accompanied by public safety events and dominate the public opinion. Meanwhile, "doubts" will lengthen the life cycle of public opinion, playing a leading role in provoking controversy. Fourthly, the Granger causality analysis is used to examine the interaction between the media agenda and the public agenda. This paper also explores the relationship between the news reports and the emotions expressed in microblogs. It is found that the media agenda influences the issues of the public agenda, while the public agenda affects the issue attribute of the media agenda. At the same time, in the "Qing'an shooting incident" situation, the emotional words used in the news reports and the emotional types expressed in microblogs has a significant positive correlation. The main improvements and innovations of this study are the following three aspects: Firstly, the news data and social media data related to the event are integrated into the evolution analysis of the network public opinion. Under the guidance of the agenda setting theory, the content level of the traditional public opinion analysis is extended to the event layer, the topic layer and the opinion layer. The agenda setting analysis is extended by transferring the issue level to the attribute level which improved the public opinion framework. Secondly, this paper improves the way to dynamically track events and uses the community detection method to automatically extract the focusing sub-event. The dynamic topic model is introduced to extract the microblogging topic, and the evolution of the topic is explored from two levels: the topic popularity and the keywords. We also put forward the method of microblogging opinion portraits and use Word2Vec to expand the sentiment dictionary. These works make up for the lack of network public opinion evolution analysis. Thirdly, it explores the interactive relationship between different levels of network public opinion under different event contexts, which provides a theoretical basis for future analysis of public opinion evolution, cluster behavior and sentiment communication.
查看全文：	预览  下载(下载需要进行登录)