特定事件情境下中文微博用户情感挖掘与传播研究

发布时间：2018-05-30 21:54

本文选题：情感分析 + 情感词表　；参考：《南开大学》2014年博士论文

【摘要】：微博等在线社交媒体在舆论的传播方面所起的作用已越来越明显,社交媒体中的用户可以自由发表他们对某一事件的观点和看法,也可以通过文字、图片、视频等形式发泄情绪。由于社交媒体中的用户处于不同的社交网络中,信息的传播非常迅速,所以在特定事件情境下,社交媒体用户极易产生群体极化现象,甚至导致网络或实际生活中的群体性事件。用户表达的情感不仅能影响事件的传播速度,而且情感能够相互感染,不利情感或负面情绪能够激发用户的负面行为,促使事件朝着不利的方向发展。所以有必要对微博等社交媒体用户在特定事件情境下的情感进行分析,判断用户的情感类型和情感极性强度,寻找影响用户情感表达和传播的影响因素,进行有针对性地监控和引导。基于以上背景,本文将研究问题集中于特定事件情境下中文微博用户的情感挖掘与情感传播研究。本研究需要解决三大研究问题,分别为情感特征识别问题、情感特征统计和描述问题和情感传播问题。主要研究工作包括：第一,构建中文情感分类词表,包括情绪分类词表和评价分类词表。词表中的情感词一方面来源于现有三个中英文情感词表,另一方面来源于待分析事件的微博文本语料,采用基于HowNet知识库标注的方法实现情感词的分类和极性强度判断。最终情绪分类词表包含12个大类和32个小类,共3773个情绪词；评价分类词表包含8个大类和100个小类,共12844个评价词；第二,实现情感词可视化和情感特征分类统计。通过情感词在同一微博中的共现计算情感词之间的关系,然后通过位置算法将情感词之间的关系通过图形的方式进行展示,同时还通过字体的大小和颜色来表示情感词的热度和极性强度。研究发现高频中心词反映了事件的主导情感类型,越靠近边缘位置,情感词越能反映普通公众的情感。对情绪词的分类统计能发现特定事件下用户所表达的各情绪类型的强度。将表情符号按正负面极性分布进行时间序列的统计可以发现一些难以发现的水军广播,去除水军广播后正负面表情强度的变化趋势相似；第三,构建特定事件信息传播网络,通过社会网络分析方法分析事件信息传播网络中的关键用户、信息传播距离、传播网络集聚程度。将用户情绪嵌入事件传播网络中,进行信息传播网络用户情感可视化,了解用户情感的分布情况。分析用户的情感表达与用户的角色之间的关系,发现决策者应关注表达“激动”、“诋毁”、“同意”或“反对”等情绪的用户,表达这些情绪的用户更容易在信息传播中起关键作用。本研究的创新点主要有三个方面：第一,完善了情感词表的构建。目前虽已有少数关于中文情感词典构建的相关研究,但一方面这些情感词表无法公开使用,少数可用的情感词表仅仅将情感词分为正面和负面两类,由于这方面的限制,目前的情感分析大多集中于对句子或文本的正负面极性进行判断,无法获知文本中具体的情感类型和情感强度。本研究将整个情感词表分为情绪分类词表和评价分类词表,不仅能实现极性的计算,而且能够实现具体情绪类型的分析。第二,将可视化技术应用于情感描述,有助于情感分析方法的完善。目前自然语言可视化较常见的是对文本标签或关键词进行标签云的展示,主要是对文本主题的可视化,而情感词可视化的研究和应用并不多见。本研究不仅仅将情感词进行可视化表示,还将情感词之间的关系和情感词的极性强度特征通过图形进行表达。另外,本研究不仅对情感词进行了可视化,还介绍了将用户情感在事件信息传播网络中进行可视化的方法,通过多种可视化技术和算法可为决策者提供更直观的用户情感信息。第三,推进了特定事件下中文社交网络情感传播研究。目前已有相关英文情感传播的研究,但这些研究多数关注用户日常交流网络的情感互动,而用户对事件信息的情感及这种情感如何在事件传播网络中进行传播和分布的研究较少,为了探索情感传播相关因素,本研究还对用户情感与用户在事件传播中扮演的角色之间的关系进行了分析。在理论研究方面,本文基于心理学的研究和HowNet本体构建了情感词的分类体系,可供后续研究作参考。在方法方面,提供了情感知识的表示方法,有助于目前情感分析方法的完善,并结合社会网络分析方法和相关性分析方法进行用户情感传播研究,有助于情感传播研究方法的完善。在实践方面,有助于政府有关部门了解公众在事件发生过程中的情感传播状况,为避免公众情感的集聚和极化,提供有针对性的信息。有助于企业或个人了解微博公众对事件的情绪反应和评价,通过公众情感扩散规律制定有针对性的应对策略。有助于公共管理部门、企业了解公众对自身服务或产品的情绪和评价,以改进自身服务或产品。
[Abstract]:Online social media, such as micro-blog, has been playing a more and more important role in the dissemination of public opinion. Users in social media can freely express their views and views on an event. They can also vent their emotions through words, pictures, video and other forms. Because users in social media are in different social networks, information is transmitted. Sowing is very fast, so in particular event situations, social media users are very easy to generate group polarization, and even lead to group events in the network or in real life. The emotions expressed by users can not only affect the speed of the event, but also affect each other, and the negative emotion or negative emotion can stimulate the negative behavior of the user. In order to promote the development of events in a disadvantageous direction, it is necessary to analyze the emotions of social media users such as micro-blog and other social media in particular event situations, to determine the emotional type and emotional polarity of the users, to find the influencing factors that affect the expression and dissemination of the users' emotions, and to carry out a targeted monitoring and guidance.
Based on the above background, this study focuses on the research of emotional mining and emotional communication of Chinese micro-blog users under specific event situations. This study needs to solve three major research problems, such as emotional feature identification, emotional feature statistics and description and emotional communication. The main research work includes: first, construction The emotional classification word list, including the emotion classification word list and the evaluation classification word list. The emotional words in the word list are derived from three Chinese and English emotional words, on the other hand, from the micro-blog text data of the events to be analyzed, using the method based on HowNet knowledge base to realize the classification of emotional words and the judgment of polarity. The classification word list consists of 12 large classes and 32 small classes, with a total of 3773 emotional words. The evaluation classification word list contains 8 large classes and 100 small classes, with a total of 12844 evaluation words; second, realizes the visualization of emotional words and the classification statistics of emotional features. The relationship between the emotion words in the same micro-blog is calculated by the emotion words, and then the location algorithm will be used. The relationship between emotional words is displayed in a graphic way, and the heat and polarity of emotion words are expressed by the size and color of the font. The study finds that the high frequency center words reflect the dominant emotion type of the event, the closer to the edge position, the more emotional words can reflect the emotions of the general public. The intensity of the emotional types expressed by the user under the specific event is found. The statistics of the emoticons according to the positive and negative polarity distribution of the time series can find some hard to find water army broadcasting, and the change tendency of the negative expression intensity is similar after the water army broadcasting. Third, the network of information dissemination for specific events is constructed and the society is constructed through society. The network analysis method analyzes the key users in the event information communication network, the distance of information propagation and the degree of network aggregation. The user emotion is embedded in the event propagation network, and the user emotion visualization is carried out, the distribution of users' emotion is understood. The relationship between the user's emotional expression and the user's role is analyzed. Decision makers should pay attention to users expressing "excitement", "denigrating", "consent" or "objection", and the users who express these emotions are more likely to play a key role in the dissemination of information.
There are three main innovative points in this study. First, the construction of emotional vocabulary is perfected. Although there are a few related studies on the construction of Chinese affective lexicon, on the one hand, these emotional words can not be used publicly. A few of the emotional words can only be divided into positive and negative two categories, because of this limitation, Most of the current emotional analysis is focused on the judgment of the positive and negative polarity of the sentence or text. It is impossible to know the specific emotional type and emotional intensity in the text. This study divides the whole emotional word list into the emotion classification word list and the evaluation classification word list, not only can realize the calculation of the polarity, but also can realize the analysis of the specific emotional type. Two, the application of visual technology to emotional description can help to improve the method of emotional analysis. At present, the more common natural language visualization is to display the label cloud of text labels or keywords. It is mainly the visualization of the text theme, but the research and application of the emotion word visualization is not much. In addition, this study not only visualizations of emotional words, but also introduces the method of visualizing user emotion in event information communication network, and through a variety of visualization techniques and algorithms for decision-makers. More intuitive user emotional information. Third, promote the study of emotional communication of Chinese social networks under specific events. There is a study of emotional communication in English, but most of these studies are concerned with the emotional interaction of the user's daily communication network, and how the user's feelings about the event information and this emotion are transmitted in the event communication network. There are few studies on sowing and distribution. In order to explore the related factors of emotional communication, this study also analyzes the relationship between the user's emotion and the role played by the user in the event communication.
In the field of theoretical research, this paper constructs the classification system of emotional words based on the research of psychology and the HowNet ontology, which can be used as a reference for subsequent research. In the way, it provides the expression of emotional knowledge, which is helpful to the improvement of the current emotional analysis methods, and combines the social network analysis method and the correlation analysis method to carry on the user's feeling. The study of sense transmission helps to improve the research methods of emotional communication. In practice, it helps the government departments to understand the public's emotional transmission in the event of the event, and to provide pertinent information to avoid the gathering and polarization of public sentiment. It will help enterprises or individuals understand the emotional response of the micro-blog public to the event and Evaluation, the formulation of targeted coping strategies through the law of public emotional diffusion. It helps the public management department to understand the public's feelings and evaluations of their own services or products to improve their own services or products.
【学位授予单位】：南开大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：G206

【参考文献】