动态网群组织CMOs建模分析与研究

CORC > 自动化研究所 > 中国科学院自动化研究所 > 毕业生 > 博士学位论文

题名	动态网群组织CMOs建模分析与研究
作者	王晓
学位类别	工学博士
答辩日期	2016-05
授予单位	中国科学院大学
授予地点	北京
导师	王飞跃
关键词	网民群体组织行为建模多智能体大数据水军
学位专业	社会计算
中文摘要	移动智能设备的普及和数据分析技术的日臻完善，促进了社会化媒体的爆发式增长，网络虚拟世界已然成为人们日常活动的重要空间，大规模网民群体组织运动(Cyber Movement Organizations, CMOs)及其行为演化研究在社会化媒体营销和共享经济中的作用愈加显著。CMOs可视为针对某一话题或事件在短期内聚集在一起，参与、讨论并共同实施某些社会行为的在线网民群体。由于CMOs成员不受现实物理世界的空间限制、组织方式多变、人员身份模糊、言辞约束宽松，且网络的放大作用明显，虚拟空间中极短时间内即可累积大量舆论及行为能量，对社会管理、经济稳定和国防安全产生重大影响。近年来，CMOs群体行为动力学及其组织演化研究吸引了众多学科领域大量学者的广泛关注。在多源融合跨领域的社会媒体平台中，网民既是虚拟资源的生产者，也是提供者、传播者和消费者。单一的意见领袖识别、影响力传播以及舆论演化模型难以准确地表达和解释社会媒体平台中网民用户交互行为的形成和组织过程，无法充分考虑人-资源(信息)-环境之间的相互作用，因而无法阐释网民提供、生产、传播和消费网络虚拟资源的动机，更无法有效说明在此过程中涌现出的复杂群体行为模式及其形成机理。人工社会方法基于自底向上的多智能体建模，可有效地应对复杂系统无法拆分还原、高度不确定、复杂多变的特征。在人工社会的基础上，王飞跃研究员提出了人工社会(Artificial Societies)、计算实验(Computational Experiments)和平行执行(Parallel )的三位一体的ACP理论框架，为动态网民群体的计算建模和实验评估提供了系统化的研究思路和解决方案。本文以复杂系统的ACP相关理论为基础，结合社会学、统计物理学、计算机仿真科学和认知心理学等交叉科学的相关技术与方法，对网络社区中CMOs参与成员个体行为的时空特性、核心群体人员的影响力评估方法、群体组织行为的动态演化、社会行为的传播机制问题进行了深入研究。同时，引入自底向上的基于多智能体建模的人工社区方法，构建社区环境，在此基础上对社区用户的交互行为机制及群体组织活动进行计算实验，揭示微观个体行为机制与宏观群体行为模式的内在联系以及群体行为有序状态的形成过程。本文的主要研究成果如下： 1. 论文提出了基于核心群体成员行为特征的CMOs规模的预测方法。该方法能够有效克服影响力模型中个体影响力随时间推移和话题变化而失效的问题。针对非社交结构化的、以内容为一级组织特征的社区论坛，首先分析了由话题引发的网民群体及其组织的互动行为统计特征；然后考虑了不同社区板块群体行为的差异性，构建由相似主题的话题所构成的主题领域空间，并提出了基于用户参与频度、参与行为密度的核心群体模型；以实验的手段揭示了核心群体对于CMOs发展的推动作用。在此基础上，基于CMOs中核心群体用户的数量进一步对CMOs的发展趋势和规模进行了预测分析。 2. 论文构建了目标引导的CMOs模型。引入粒子群的思想对由核心群体推动及目标引导的CMOs组织及其演化进行了设计、建模和分析，从社会学习和心理认知的角度解释了微观个体行为机制，及其与宏观群体行为模式的关联关系。 3. 论文提出了网民交互行为的人工建模与计算实验方法。基于自底向上的多智能体建模方法，构建在线论坛社区的人工社区模型，以此为基础提出了用户的话题发布机制和评论机制，并对其展开计算实验分析及验证。使用大规模真实社会媒体平台数据为输入，对宏观层面上的网民行为特征进行统计分析和抽取，搭建人工社区环境。通过对个体特征参数的调整，对社区环境中群体行为的幂律分布特征展开实验，分析并计算了个体内在属性以及社区环境对于个体行为决策的影响，进而揭示了网民在群体行为组织演化过程中的行为决策机制。 4. 论文设计了水军智能体及其任务执行的模型。以人工社区环境为基础，对水军这类特殊的CMOs进行任务、活动及策略建模，针对水军对话题的干涉行为展开计算实验。在引入累积暴露风险指数对水军干涉行为进行约束，以及舆情极性阈值量化水军行为目标的基础上，考虑两类CMOs(水军和普通用户)的协同竞争演化，设计水军智能体对普通用户智能体的“说服”任务，基于人工社区环境对水军的干涉行为进行计算实验，分析并量化水军不同干涉策略的有效行为转化率及其影响效果。综上，本文围绕社会化媒体中网民行为管理需求，针对大规模网民行为及其组织演化的建模问题展开深入探讨和研究。综合运用ACP理论、社会学习理论、社会网络分析、多智能体建模等方法，考虑网络世界中人-资源-环境之间的交互及协同演化问题，从实证分析、仿真建模和计算实验三个角度，重点对个体行为机制、单方群体组织演化、对立群体协同竞争展开研究。本文研究将为社会应急管理、企业的社会化营销服务以及国家安全监督提供合理有效的理论依据和决策支持。关键词：网群运动组织; 多智能体建模; 人工社区; 计算实验
英文摘要	Recently, the pervasive use of portable intelligent mobile devices and the improving data analysis technologies have greatly accelerated the development of social media sites. Cyberspace is now an important part of people’s living space. It brings new opportunities to studying human behavioral dynamics with Big Data in a global and systematical view. The studies of Cyber Movement Organizations (CMOs) and their applications in social marketing and sharing economy are becoming more and more significant. Many related researches focus on studies of opinion-leader identification, influence diffusion and behavior infection in social networks. However, with the rapid development of all kinds of social media sites (SMS), online users with various education backgrounds, occupations and ages are attracted, and resources from multi-sources across domains are generated, resulting to the uncertainty, variety and complexity of the dynamics of social networks. Thus, classical models considering factors separately are neither suit for explaining the organization of cyber behaviors, nor suit for illustrating the motives of users’ interactive behaves. Besides, the connection between individual behavior mechanism and collective behavior pattern is rarely studied. The ACP approach, integrated by artificial societies (A), computational experiments (C) and parallel execution (P), works efficiently to deal with the uncertainty, variety and complexity of social systems using multi-agent modeling, and provides systematized thought and solution for the computational modeling and experimental evaluation of CMOs. This paper makes use of the ACP approach, combining with related technologies and methods in sociology, statistical physics, simulation science, social movement organizations .etc, to explore the spatio-temporal behavioral features of individuals, evaluate the importance of core community members, study the organizational evolvement of cyber behavior and verify the social contagion mechanisms on social media platforms. Cyber behavioral motives are analyzed considering effects from two sides: social influence from other individuals, and individual cognition from people’s education, family, work, .etc. The multi-agent modeling method is introduced to construct an artificial community, in which the cyber interactive behaviors and organizational evolvement can be studied with modeling and experimental leverages. Based on these, computational experiments are conducted to reveal the inner connection between individual behavioral mechanism and collective behavioral patterns. The major works and achievements consist of four aspects: A method of predicting the scale of CMOs in online forums based on core group members is proposed. Online forums are nonsocial structured, while the cyber movements inside are organized around topics. A theme domain space (TDS) is constructed crossing boards by topics of the same theme, which clusters users of similar interest. Thus, the statistical characteristics of cyber behavior are studied inside of a targeted TDS. On this basis, the concept of core group members, who show high participate-frequency, short participation-interval and long life cyber characteristics, is put forward. Compared with the IDM model, the proposed method achieves a better accuracy rate of influence measure, especially in identifying leaders in the entire theme domain space. Furthermore, the studies on predicting CMOs’ size based on core group members are conducted. The thoughts in particle swarm optimization are introduced to illustrate the gathering and organizing process of cyber collective behaviors. Using social learning and individual cognitive theories, the individual behavior mechanism and its connection to global behavior patterns are illustrated. The analysis of cyber interactive behaviors using artificial communities and computational experiments are implemented. The relationship between the position of topics and their ability to attract comments is quantified through statistical analysis of 10 years data. The multi-agent modeling method is implemented to construct an artificial community from the bottom up, in which the individual’s post-publishing mechanism and the comment-making mechanism are designed considering social and individual impact factors. The emergence of global patterns from micro-level interactions is studied by computational experiments, and further analysis reveals the decision mechanism of individuals in CMOs. An agent model of astroturfer is developed, and the persuasion process of astroturfers to common users is explored. The CMOs of astroturfers and their tasks are modeled in the constructed artificial community with experimental leverages. Interacts between astroturfer-agents and targeted users are modeled in accordance with actual social marketing situations. A factor named cumulative exposure risk index is introduced to avoid exposing the real identity of astroturfers. The effect and impact of different behavioral strategies of astroturfers facing common users with different properties are tested by implementation of computational experiments. The method considers the limitation of existed methods of empirical analysis and theoretical derivation models. It shows great potential value in social marketing. This paper considers the management needs for cyber behaviors in social media sites, explores and discusses the organizational evolvement of large-scale cyber behaviors. The ACP approach, combining with social learning theory, social network analysis and multi-agent modeling methods, is used to comprehensively study the interactive and collaborative evolvement of users, information and social media platforms. The research results provide social emergency management, enterprise marketing management and national safety management with efficient data and technology supports. Keywords: Cyber Movement Organizations; multi-agent modeling; artificial communities; computational experiments; astroturfers
语种	中文
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/11597]
专题	毕业生_博士学位论文
作者单位	中国科学院自动化研究所
推荐引用方式 GB/T 7714	王晓. 动态网群组织CMOs建模分析与研究[D]. 北京. 中国科学院大学. 2016.