复杂系统的宏观性质和行为取决于系统中个体之间的相互作用结构，而这一结构可以通过复杂网络来描述，所以复杂网络研究就成为探讨复杂性的基础。其中社会系统，是复杂网络平台下的一个重要研究对象。社会系统中的集群动力学，如舆论形成、行人动力学、人口迁移得到了越来越多社会学家和物理学家的共同关注。其中舆论形成研究思想，起源于哈肯的协同学，认为舆论是系统内部发生的某种相变，是系统各个部分发生了相互协作，使系统涌现出微观个体层次上不存在的新的结构和特征。之前的研究主要集中在网络的结构对舆论形成的影响。越来越多的证据表明，舆论的形成是网络拓扑结构和动力学共同作用的结果。局域作用、社团结构以及空间结构是社会网络的重要特征，结构因素和动力学因素如何影响舆论形成从而推动舆论形成成为目前舆论形成的核心问题之一。围绕这一核心问题，本文就局域信息、社团结构和空间结构在舆论动力学中的地位和作用展开了研究。首先，提出了基于局域作用的自适应舆论模型，探讨了局域因素在结构和行为自适应的舆论模型中的影响和作用。此局域模型主要包含两个部分，局域类聚过程和局域影响过程。发现局域类聚半径越大，即局域个体能够选择结识的同类越多，则舆论越倾向于形成规模越大的舆论团体。当考虑局域影响过程时，发现局域影响半径、平均度相变点呈现出正相关关系；相变点处，终态舆论团体规模服从幂律分布。上述结果表明，局域因素在理论上舆论形成中的相变点的确定及相变行为有着重要的影响。其次，考察了社团结构对舆论形成的影响。先发现社团结构越清晰，达成舆论共识的时间越长。接着通过研究局部和整体的舆论形成，发现在社团结构清晰的网络上，社团内部比整体要更快的达成舆论共识；社团结构变模糊时，社团内部和整体几乎同时达成共识。这表明清晰的社团结构会阻碍共识的达成，说明内在的社团结构在舆论形成的过程中有重要作用。再次，基于“社团结构清晰会延长舆论传播时间”的结论，探讨了改变网络社团结构的途径和效果。结合以往划分社团结构中采用的局域信息分裂算法（边集聚系数算法）的思想，提出了分别基于集聚系数和三角形等的改变网络社团结构的途径。并利用平均边集聚系数，经典模块函数，以及相异性函数，分析了三种扰动方式对网络社团结构的影响，从改变网络社团结构的角度，为缩短舆论传播时间提供了新的方法和途径。最后，研究了空间结构对舆论形成的影响。在具有空间结构的社会网络上，探讨了空间无标度对自适应模型的影响。实证表明，很多实际网络的空间距离服从幂指数为“-1”的幂律分布，在空间网络上模拟舆论动力学，发现在幂指数为“-1.2”处，舆论达成共识最快。基于此，提出了基于空间结构的的自适应舆论模型。在类聚过程中，考虑空间距离的成本的影响。类聚的对象不仅限制为同类，而且距离服从幂指数分布，即距离越大，类聚的概率越小。通过调节类聚空间幂指数，使网络的空间距离呈现幂律分布特征。本小节的探讨，主要弥补了目前空间结构上舆论形成研究的不足。本文从局域作用、社团结构、以及空间距离无标度性质等各个方面对舆论形成进行了深入细致的研究。通过建立局域自适应模型，初步探讨了了局域因素在舆论形成中的作用，发现了局域同质相连导致了大集团的湮灭，局域影响半径和平均度决定了舆论从大规模整体共识变为小团体局部共识的相变点；通过研究改变社团结构对舆论动力学的影响，发现社团结构越清晰，舆论传播时间越长，初步探讨了社团结构在舆论形成中的作用；通过改变社团结构的途径和效果,发现了改变社团结构更有效的方式，为缩短舆论传播时间提供新的方法和思路；探讨了空间无标度性质对舆论传播时间的影响, 发现在空间距离幂指数为“-1.2”时，舆论传播最快；并构造了基于空间结构的自适应舆论模型，可使网络的空间距离分布呈现幂律特征。通过讨论舆论形成机制，以及舆论传播时间等动力学行为，进一步说明了局域作用、社团结构以及空间性质在舆论形成中的地位和作用。

The macroscopical properties of complex system lie on the interaction structure between individuals, which can be described by complex networks. Thus, the study of complex networks becomes the foundation of complexity. Social system, is an important research object under the framework of complex network. The collective dynamics in social system, for instance, Opinion formation, Herd Behavior, Pedestrian dynamics, and Segregation phenomenon, draws more attention from both socialists and physicists. The study of opinion formation originated in Synergetics proposed by Haken. Haken suggested that opinion formation is a phase transition and new structure and character caused by cooperation of sub-systems. Former studies mainly focus on the effect of variation of network’s structure on opinion formation, however, more evidence shows that opinion formation is a co-evolution process where network’s structure and dynamical behavior are adaptive to each other. Until now, the key factor in opinion formation and the mechanism opinion formation become the key problems in study of opinion formation. Based on these key problems, opinion formation is particularly studied in this thesis from following aspects, including the role and status of local information, community structure, and spatial structure in opinion’s formation model.Firstly, local adaptive model of opinion formation is proposed and the role and status of local factor are investigated in chapter 3. Local adaptive model consists of two parts: local homophily process and local influencing process. As the radium of local homophily process increases, every individual tends to acquaint more like-minded persons and large community holding on the same opinion will be formed. Meanwhile, it is found that the radium of local influencing and the average degree of network are positive to the transition point. The results in chapter 3 illustrates that local information is important in opinion formation theoretically.Secondly, the effect of community structure on opinion formation, especially convergence time, is investigated in chapter 4. While clear community structure of arbitrary network becomes obscure, the convergence time to consensus state is much faster. Through investigating local and global opinion formation, local consensus is found to be reached more quickly than global consensus in network with more clear community structure. In network with obscure community structure, local consensus is formed almost at the same time with global consensus. This result implies that significant community structure hinders the formation of consensus in some extent and the intra community structure plays important role in opinion formation.Thirdly, based on the relation between the significance of community structure and efficiency of opinion spreading introduced in chapter 4, the approach and effect of changing network’s community structure are discussed in chapter 5. Methods based on the clustering coefficient to purposely perturb network’s structure are proposed. Through the measurement of clustering coefficient, modularity function, and dissimilarity function, the variation of community structure is analyzed. Lastly, how spatial structure of network influences opinion formation is investigated in chapter 6. A classic opinion dynamics is simulated on spatial network, and the relation between spatial character and convergence time is discussed. It is found that system can reach steady state using the least time while at certain power-law exponent “-1.2” of distribution of spatial distance. Then an adaptive opinion model based on spatial structure is proposed, and distribution of distance in network obeys to power-law at certain parameter. In this chapter, the spatial factor shows powerful effect on opinion formation.In conclusion, opinion formation is investigated in this thesis from local information, community structure and spatial structure’s effect. The local adaptive model is constructed and it is found that transition point depends on the radium of local influencing and average degree of network. The relation of significance of community structure to the efficiency of opinion spreading is found to be negative, and then the approach and effect of changing community structure are investigated. Lastly, opinion spreading is found to be the most efficient in scale-free spatial network at certain pow-law exponent. Inspired by spatial structure in social network, the adaptive opinion model based on spatial structure is proposed. The role of local interaction, community structure, and spatial structure in opinion formation is deeply studied by mechanism and dynamical behavior in opinion formation.