【作者】 王晓；

【导师】 李纲；

【作者基本信息】 武汉大学 ， 情报学， 2019， 博士

【摘要】 随着移动互联网技术的快速发展,社交媒体平台在公众议题舆论生成和发酵演化的过程中的影响地位日渐突显。面对舆论环境的重大变化,无论是政府还是企业,对网络舆论的监控和引导的需求都愈发紧迫,并提出了更高要求。同时,随着网络媒体与传统媒体不断推进的深度融合、用户对于网络舆情环境的逐步适应和理性回归,以及政府机构对于网络舆情的深入认识和规范管理,使得意见领袖群体在经历了一系列发展过程之后得以自然淘汰和清理整顿;留存下来的往往具有更深厚的知识和经验积累、更专业的信息素养、更强烈的责任意识,因此在网络舆论中也具有更高的影响能力和媒体价值。是以,通过对意见领袖的舆情关注偏好和观点表达倾向展开研究,一方面有助于加强社交媒体中关键意见群体的管理,另一方面有助于完善网络舆论管理、疏导民意合理表达。然而,现有相关研究中,研究成果大量集中于意见领袖识别且实验主要在新浪微博中开展,主题分布的分析内容尚停留在基础表面,同时,持续性主题发现缺乏明确定义和量化方法。基于此,本文针对意见领袖主题分布和持续性主题发现展开研究。全文共分为七个章节,所包含内容如下。第一章,引言。基于本文研究背景、研究意义的阐明,随后对意见领袖、主题提取与分布、持续性主题发现、观点挖掘与漂移的国内外研究现状进行了全面的梳理和总结。进而明确本文研究思路,绘制本研究的技术路线图。基于此,详细介绍本文研究内容,归纳研究创新点。第二章,概念界定与理论基础。首先,界定了相关概念的具体表述,包括意见领袖、意见领袖主题、持续性主题和观点。其次,探讨了相关理论基础,对“把关人”理论和“议程设置”理论的基本描述定义、模式、特点、传播效果及其在网络环境中的表现变化进行了分析。最后,阐述了相关工具方法,对意见领袖识别和主题提取的主要方法进行分析。第三章,意见领袖识别及其主题提取。从识别流程、实验平台、舆情事件和识别指标四个方面构建意见领袖识别方法体系;继而围绕着新浪微博和知乎关于“江歌案”的讨论,进行了意见领袖识别实验,并从基础比较、角色定位和互动强度三个层面对识别结果展开分析。进而,从提取流程、数据获取与预处理、文本向量表示和文本主题挖掘四个方面明确了意见领袖主题提取方法,基于所识别出的新浪微博和知乎意见领袖为期三年的发文文本数据进行主题提取实验,并在筛选得到主题词的基础上,参考多个社交媒体和新闻网站内容,确定主题标签和分类。第四章,意见领袖主题的三维分布。首先,从时间、空间、人际三个维度阐释本文所提出的意见领袖主题分布分析框架,并对量化方法和判断标准进行了界定。其次,分别从整体、主题和意见领袖三个角度,对知乎、新浪微博个人和新浪微博机构意见领袖主题的历时分布展开分析。接着,分别从领域层面和主题层面,对这三组意见领袖分别代表的特定网络活动空间中主题的分布广度、集聚程度,以及不同空间两两之间的关联程度展开分析。最后,一方面,由意见领袖的主题耦合揭示意见领袖之间基于相同的讨论兴趣所建立的关系;以意见领袖为节点、主题耦合为边、主题耦合强度为权重构建关系网络,分析意见领袖的兴趣相关性。另一方面,由意见领袖的耦合关系揭示主题之间基于相同的讨论参与者所存在的隐性关系;以主题为节点、意见领袖耦合为边、意见领袖耦合强度为权重构建关系网络,分析主题的隐性关联。第五章,持续性主题和主题簇发现。首先,依据主题运行过程分析和主题相对讨论强度的计算方法,给出了本文对于持续性主题的明确定义。其次,基于主题相似性计算与过滤和主题簇提取条件,制定了主题簇发现方法。最后,结合实际数据,进行了持续性主题发现、主题相似性计算与过滤、主题簇的发现及持续性判断三项实验,提取出知乎和新浪微博中的持续性主题和主题簇。第六章,持续性主题中的观点漂移。首先,在将意见领袖的持续性主题发文中表现出随时间推移产生的观点变化进一步细分为:观点情感倾向和观点评价对象两个部分的基础上,分别阐释了观点情感倾向分析和观点评价对象分析的计算流程,对其表示形式和计算方法进行了设计,并界定其漂移检测的判断条件。其次,对持续性主题中观点情感倾向的整体分布和漂移变化进行分析,并比较知乎和新浪微博中表现的异同。最后,对持续性主题中观点评价对象的整体侧重和漂移变化进行分析,同样比较知乎和新浪微博中表现的异同。第七章,研究总结与展望。总结本文研究结论,并指出存在的不足和未来的研究方向。更多还原

【Abstract】 With the rapid development of mobile Internet technology,the influence of social media platforms in the process of public opinion generation and development on public issues has become increasingly prominent.Facing with major changes in the public opinion environment,both the government and the enterprise put forward an urgent need and higher requirements for monitoring and guiding the network public opinion.Meanwhile,with constantly deeper integration of online media and traditional media,users conform themselves to the internet environment and rational return of thinking,as well as standardized management by governmental agencies through their deeply understanding on online public opinion,the opinion leader group has experienced a series of development process and be cleared up and reorganized.The retained part often has deeper knowledge and experience accumulation,more professional information literacy,and stronger responsible consciousness,therefore,possessing higher influence ability and media value in online public opinion.Hence,an in-depth study of text content posted by opinion leaders should help strengthen the management of key opinion groups in social media on the one hand,while improve the management of public opinion and rationalize public opinion on the other hand.However,in the existing research,the productions are concentrated in the identification of opinion leader and most of the experiments are carried out in Sina Weibo,the analysis of subject distribution stays on basic layer,and the durative subject lacks clear definition and quantitative method.Based on this,this paper conducts a study on the subject distribution and durative subject discovery of opinion leader.The full text is divided into seven chapters,the main contents are as follows.Chapter one is the introduction.After clarifying the research background and significance of this thesis,the domestic and international research status of opinion leader,subject extraction and distribution,durative subject discovery,opinion mining and drift are combed and summarized.The research thoughts are then clarified.With technical roadmap be drawn,the research content is stated in detail and major innovative points are summarized.Chapter two is Related concept definition and theoretical basis.First,clarify the definition of basic concepts,including opinion leader,opinion leader subject,durative subject and opinion.Next,with reference to the "gatekeeper" theory and the "agenda setting" theory,introduce its definition,model,characteristics,communication effects and its performance changes in the network environment.Finally,the related tools and methods are described,analyze the main methods of opinion leader identification and topic extraction.Chapter three is opinion leader identification and subject extraction.Identification method of opinion leader is defined from four aspects,which are identification process,experimental platform,public opinion events and identification indicators.Then,based on the data of discussion about “Jiangge case” in Sina Weibo and Zhihu,opinion leaders are identified.The identification results are analyzed from three parts,including basic comparison,role positioning and interaction intensity.Next,the opinion leader subject extraction method is defined from four aspects,which are extraction process,data acquisition and preprocessing,text vector representation and subject mining.The subject extraction experiment is accordingly conducted on the three-year posts data of the identified opinion leaders.Based on the screening of keywords,and referring to the content of multiple social media and news websites,tags and classifications of the subject are determined.Chapter four is the three-dimensional distribution of opinion leader subjects.Firstly,interpret the distribution analysis framework of the opinion leader subject proposed in this paper,including three dimensions of time,space and interpersonal,define the quantitative methods and judgment criteria as well.Secondly,from the perspective of the whole,the theme and the opinion leader,analyze the diachronic distribution of opinion leader subject from different groups named Zhihu,Weibo personal and Weibo organizational.Thirdly,from the domain and subject level,analyze the degree of coverage and concentration of subject in the different cyberspaces which is respectively represented by three groups of opinion leaders,and the degree of association between different cyberspaces.Finally,on the one hand,the subject coupling of opinion leader reveals the relationship between opinion leaders based on shared discussion interests.By constructing a relational network with opinion leaders as nodes,subject coupling as edges and the coupling strength as weights,the interest correlation between opinion leaders is analyzed.on the other hand,the implicit relation between subjects,which is based on shared discussion participants,is uncovered by opinion-leader coupling.Take subjects as nodes,opinion-leader coupling as edges and the coupling strength as weights to construct a relational network,and analyze the implicit association between subjects.Chapter five is discovery of durative subject and subject set that comprises several similar subjects.Firstly,according to the subject running process analysis and the calculation of subject relative discussion intensity,a clear definition of the durative subject is proposed.Secondly,based on the subject similarity calculation and filtering and subject set extraction conditions,the subject set discovery method is developed.Finally,applying to actual data,three experiments that are durative subject discovery,subject similarity calculation and filtering,subject set discovery and durative measure,are carried out to extract durative subject and subject set in Zhihu and Weibo.Chapter six is opinion drift in durative subject.By subdivided the opinion changes generated over time in the posts of durative subject and subject set into two parts: opinion sentiment tendency and opinion evaluation object,firstly,expound separately the calculation process of opinion sentiment tendency analysis and opinion evaluation object analysis.While also clarify their representation form and calculation method,and the judgement condition of drift detection.Secondly,analyze the overall distribution and drift of the opinion sentiment tendency and opinion evaluation object in durative subject,and compare the performance in Zhihu and Weibo.Finally,analyze the overall distribution and drift of the opinion sentiment tendency and opinion evaluation object in durative subject set,interrelate and compare the performance with what in durative subject of the same platform.Chapter seven is research summary and respect.Summarize the conclusions of this paper and point out shortcomings and future research directions.更多还原