【作者】 王洁；

【导师】 孙华；

【作者基本信息】 山东大学 ， 管理科学与工程， 2023， 硕士

【副题名】基于言语行为理论视角

【摘要】 随着互联网的发展,人们对知识的需求不断增加,用户生成内容的方式越来越受欢迎。在此背景下,问答社区逐渐成为人们寻找解决问题答案的重要手段。在社交问答社区中,不同类别的线程中存在不同的知识共享活动和用户交互。以往的文献大多集中在在线互动过程中用户的感知与体验,或关注答案和问题质量的影响因素,或研究成员之间社交网络的特征发现用户互动的特点,未对每个讨论中的实际对话过程进行更深入的探讨。而在问答社区中,成员之间的绝大部分社交互动都是基于线程的文字互动,但目前研究大都关注单条帖子或整个社交网络,缺乏对线程层面言语互动的结构分析;此外,用户间言语和信息的传递也主要依赖于文字内容,它能够直观地反映用户的意图和目的,而目前关于问答社区内容方面的研究主要集中在简单的话题分析或语义关联、情感分析等,少有研究从言语行为视角探究用户间是如何进行沟通交流的。了解线程层面的言语互动有助于促进用户对话,进一步帮助问题的解决,维持社区的健康发展。为此,本研究基于言语行为理论,为每个帖子人工分配对话行为标签,研究不同类别线程的对话结构差异;并构建对话网络,探究不同对话网络变量对问题解决度的影响。具体来说,本文基于CSDN社区爬取数据集,以252条线程为研究对象,使用过程挖掘和频繁子图挖掘技术,揭示满意结帖和无满意结帖线程的结构差异。此外,以对话结构的网络指标为自变量,构建影响问题解决度的计量模型,使用有序Logistic回归方法进行假设检验。本研究结果发现,满意结帖线程往往比无满意结帖线程有更多与答案相关的帖子、更少与问题相关的帖子,且在满意结帖线程中,包含更多的文明式参与和互动。关于实证研究部分显示,图密度、平均路径长度和Q/A比率对问题解决度具有显著负向影响,而平均聚类系数对问题解决度具有显著正向影响。本研究在理论上有助于从内容和网络两个角度探究社会问答社区的知识共享过程,在实践方面可以为社区用户和管理者提供建议。本研究的创新之处主要包括两方面。首先本文以线程为基本研究单位,过去的文献大多集中在对单条帖子或者从社区整体层面进行分析,少有研究检查线程级别对话结构的影响。其次,基于言语行为理论,通过比较满意和无满意结帖线程在对话结构上是否有差异,进一步从对话网络的角度建立计量模型探究对问题解决度的影响。将内容与网络分析相结合,丰富了现有互动研究的内容,为后续研究提供新的研究视角。更多还原

【Abstract】 With the development of the Internet,people’s demand for knowledge is increasing,and user-generated content is becoming more and more popular.In this context,Q&A communities have gradually become an important way for people to find answers to their problems.In social Q&A communities,there are different knowledge sharing activities and user interactions in different categories of threads.Most previous literature has focused on users’ perceptions and experiences during online interactions,or on the factors influencing the quality of answers and questions,or on the characteristics of social networks among members to discover user interactions,without exploring the actual dialogue process in each discussion in more depth.In Q&A communities,the vast majority of social interactions between members are thread-based textual interactions,but most current research focuses on posts or the entire social network,lacking structural analysis of thread-level verbal interactions;in addition,the transmission of speech and information among members also relies heavily on textual content,which can intuitively reflect users’ intentions and purposes,while current research on the content aspects of Q&A communities focuses on simple topic analysis or semantic association and sentiment analysis and few studies have explored how users communicate with each other from a speech act perspective.Understanding verbal interactions at the thread level helps facilitate user conversations,further aiding problem solving and sustaining a healthy community.Based on speech act theory,this study manually assigns dialogue act labels to each post to investigate the conversation structure differences of different types of threads;and constructs conversation networks to explore the effects of different conversation network variables on problem solving degree.Specifically,based on the CSDN community crawl dataset with 252 threads,this thesis uses process mining and frequent subgraph mining techniques to reveal the structural differences between satisfied-closed and unsatisfied-closed threads.In addition,the network indicators of conversation structure were used as independent variables to construct an econometric model affecting the degree of problem solving,and the hypothesis testing was performed using ordered logistic regression method.The results of this study found that satisfied closed threads tend to have more answer-related posts and fewer question-related posts than unsatisfied closed threads,and contain more civil engagement and interaction in satisfied closed threads.The section on empirical studies shows that graph density,average path length,and Q/A ratio have a significant negative effect on problem solving degree,while the average clustering coefficient has a significant positive effect on problem solving degree.This study is theoretically useful to explore the knowledge sharing process of social Q&A communities from both content and network perspectives,and practically can provide suggestions for community users and managers.The innovations of this study include two main aspects.Firstly,this thesis takes threads as the basic unit of study,most of the previous literature has focused on the analysis of posts or from the community level as a whole,and few studies have examined the effect of thread-level conversation structure.Second,based on speech act theory,we further explore the impact on problem solving degree from the perspective of conversation networks by comparing whether there is a difference in conversation structure between satisfied and unsatisfied closed threads by building an econometric model.Combining content and network analysis enriches existing interaction research and provides new research perspectives for subsequent studies.更多还原