社交机器人已逐渐成为社交媒体中重要的活动主体，它们模仿人类行为并与人类进行交互，在影响社会舆论的同时也改变了传统社会中以人类为主导社会交往模式。目前，社交机器人的研究视角主要强调其所代表的技术系统特征，以算法技术将社会关系“工具化”为视角开展研究，将社交机器人视为权力的代理，对人类主体性产生干预。本文聚焦于从嵌入社会情境的人机交往行为，来讨论社交机器人所代表的动态权力是如何持续、动态地影响社会互动。通过对新型冠状病毒大流行期间，推特平台上人们关于疫苗接种的讨论进行大数据分析，本文试图回答如下两个问题：社交机器人是否能适应与影响社群变化以及社交机器人能否产生持续性影响。疫苗犹豫一直是十分重要的公共卫生问题，新型冠状病毒大流行期间人们原先的生活方式都受到了极大的影响。如今，疫苗接种已成为各国试图结束疫情的重要手段。尽管在政治与经济的压力下，全球平均疫苗接种覆盖率已达到2/3，但科学与健康问题的政治化，使得人们围绕疫苗的必要性展开了更加广泛的争论，大量社交机器人也参与其中。议题参与的广泛性、社交机器人的活跃性为本文开展人机交往的研究提供了良好的环境基础。本文主要从三个方面对社交机器人交往行为特征展开分析：语言的情绪特征，语言的内容特征以及网络的结构特征。情绪特征在社交媒体的信息传播过程中扮演着十分重要的角色，负面情绪的传播往往被视为群体极化的重要诱因。内容特征则代表信息的吸引力，推文作为信息传播的载体，其核心在于向潜在信息接收者传递了什么信息。网络结构反映出人们在交往中自组织行为的形成，以及人们在社群内外观点与行为的差异。研究首先通过聚类算法划分的意见社群，发现社交机器人在社群内外的活动中存在差异。在社群间的传播中，社交机器人更加倾向于放大消极情感，而在社群内社交机器人较少会放大消极情感，并会减弱消极情感。社交机器人在意见更为统一的社群内更加占据传播的优势地位，并且其社群地位具有累积性，单纯的情感特征不足以驱动内容的传播。在语言的特征上，社交机器人能综合的运用描述性和统计性特征来激发其他用户的反馈。在支持疫苗的社群中，社交机器人会使用政治化主题、统计性用语特征并常常与喜悦与愤怒的情绪联系；在反对疫苗的社群中，则更多的运用人物和疾病相关词语并配合恐惧、反感的词语。通过不同的语言特征，社交机器人可以更好的引起对方的回应。社交机器人能够保持长期的活跃性，并舆论改变其传播观

In social media, social bots are gaining prominence rapidly. They imitate human behavior and interact with humans, which not only influences public opinion but also alters the paradigm of conventional communication, which is dominated by interpersonal interaction. Prior research on social bots has emphasized the technological system aspects, conducted from the perspective of algorithmic ＂instrumentation＂ of social relations, and viewed social bots as interfering agents of power. This research examines how social bots are integrated in social situations and how they continuously and dynamically participate in social interactions using the dynamic flow of power represented by human-computer interaction.This research sought to answer the following questions by analyzing the vaccination-related conversations on Twitter during the current coronavirus outbreak using big data methodology. Whether social bots can adapt to and influence community, as well as whether they can have a permanent impact.Vaccine hesitancy is a developing concern for public health. The emerging coronavirus pandemic has a substantial effect on people‘s daily lives. Vaccination has evolved into an indispensable weapon for countries striving to end the epidemic. Despite political and economic constraints, the global average vaccination rate has exceeded two-thirds. With the politicization of science and health issues, however, the vaccination debate has expanded, and a large number of social bots have participated in the debate on vaccine-related topics for various reasons. The scope of subject engagement and the activity of social bots provide a robust environmental foundation for the study presented in this paper.Primarily, emotional, entity, and network structure are associated with the communication features of social bots. Emotional states play a crucial role in the dissemination of information on social media, and the propagation of negative emotions is generally considered as a primary contributor to group division. The attractiveness of information is defined by the characteristics of its content. The essence of tweets as a mechanism of spreading information is the information communicated to potential recipients. The network structure describes the emergence of self-organizing communication as well as the existence of varied perspectives and behavior within and beyond the community.The research begins by dividing opinion communities using a clustering method, and then discovers that the behavior of social bots within and outside the community differs. In inter-community communication, social bots are more likely to amplify negative emotions, whereas within a community, social bots are less likely to amplify negative emotions and are more likely to reduce bad feelings. In communities with more unified ideas, social bots play a more dominant role in communication, and their community status is cumulative; yet, basic emotional qualities are insufficient to drive the spread of content. Social bots can utilize a combination of descriptive and statistical linguistic traits to solicit feedback from other users. In the pro-vaccine community, social bots employ politicized themes, statistical linguistic traits, and are typically connected with sentiments of happiness and wrath; in the anti-vaccine group, more disease-related phrases and coordinated Fear, disgust words are used. Social bots are better able to elicit responses from each other through the use of a variety of linguistic elements. Social bots can be active for an extended period of time, and public opinion can alter its propagation views - social bots increase the diversity of dissemination during periods of active public opinion and concentrate more on current opinions during periods of tranquil public opinion.Overall, social bots are better able to grasp the differences in community structure, not only to stimulate the community‘s perspective from the periphery, but also to maintain community stability within the community. Using the method of big data analysis, this study analyzes whether social bots may employ emotional and linguistic characteristics to impact the transmission of information and whether social bots can demonstrate changes in network topology in terms of information dissemination. On the basis of this premise, it is evident that social bots shape the freedom to speak by regulating the interaction between individuals and groups, and that the ＂instrumentation＂ and ＂objectification＂ of social interactions do not cause for limiting and restricting individual choices.The algorithm of social bots derives its strength from the development of cooperative actions. Social bots must facilitate individual autonomy in order to shape group actions. Otherwise, regardless of the driver‘s level of authority, the social bot‘s actions cannot transmit this power to the individual. We must acknowledge that social bots may enable a new form of social integration. Instead of confining the induction and control of individuals to a restricted set of options, it studies the diversity of individual social interactions via adaptive and dynamic characteristics.