社交媒体正在加剧在线意见极化，很大程度上，“极化”始终伴随着“沉默的螺旋”。聚焦于这一过程机制，借助基于主体的建模与仿真，该研究层层递进地关注了网络语境、网络中的群体构成以及网络结构特征的具体影响，并尝试探索了“社交媒体倦怠”的影响，从而详尽探究了个体在网络中进行意见互动时如何调整表达或沉默意愿，这些个体微观层面的行为又如何诱发了群体宏观层面“沉默的螺旋”与“极化”。该研究的概念模型中划分了四组建模主体：普通个体、中坚分子、意见领袖、媒体，其“意见值”、“表达意愿”、“在线关系网络”属性迭代更新规则、行为交互规则均有所差异。针对四个核心研究问题，共计设计了12种实验场景，并利用NetLogo软件实现了仿真。此外，也对模型进行了充分的有效性评估。围绕网络语境：（1）100次迭代仿真结果揭示了意见互动次数的稳态特征——无论是否存在网络语境，个体平均在经过10次感知意见气候、媒体气候后，“沉默的螺旋”与“极化”现象基本不再反复；（2）相较而言，“沉默的螺旋”现象更易发生于无网络结构时。在对照实验场景中，网络语境确实削减了“沉默的螺旋”，但伴随着“极化”加剧。围绕网络中的群体构成：（1）研究发现意见领袖的作用反而不及中坚分子。这类“沉默的反抗者”可以逆转“沉默的螺旋”，并加剧意见“极化”；（2）在仿真“媒体干预”机制时，“极化”的削减伴随“沉默的螺旋”加剧，始终持中立意见的群体几乎不受影响；（3）在媒体的意见值或表达意愿不再恒定的两类场景下，均存在“沉默的双螺旋”现象，媒体与其他群体形成此消彼长的双向互动，媒体过度“沉默”与“反转”均会造成在线意见互动的失衡。围绕网络结构特征：（1）分别模拟演绎了无标度网络、小世界网络。个体在前者中更易保持沉默，通常体现为孤立节点，且其所持意见基本属于同一社区；（2）另仿真了两种网络更新机制——在线寻找潜在好友、在线寻找意见社区，结果表明：后者中的“偏好连接”是导致“沉默的螺旋”与“极化”的主要助推剂，与此同时也可以观察到“回音室效应”。进一步地，围绕“社交媒体倦怠”：仿真发现在线意见互动群体过于活跃或过于沉闷时，都会导致舆论极化。一方面，活跃的在线关系网络更新有助于抑制“沉默的螺旋”，使舆论具有背反的过程。然而，消极意见与此同时也更易传播；另一方面，在“社交媒体倦怠”过强时，舆论易由单一化的意见主导，沉默与极化之中或存在网络舆论与现实舆论的剥离。
Social media is exacerbating the polarization of online opinion, which is largely accompanied by the "spiral of silence" phenomenon. Focusing on this mechanism, this study used Agent-Based Modeling and Simulation (ABMS) to progressively examine the specific impact of the online context, the group composition (the role of opinion leaders, hardcore, and media), and the network structure. The potential impact of the variable "social media fatigue" was also focused on. In this way, this study comprehensively investigated how individuals adjust their willingness to remain expressive or silent when interacting online, and how these individual behaviors at the micro level trigger the “spiral of silence” and “polarization" at the macro level.Based on the literature review, four groups of agents were identified in the conceptual model of this research: netizens, hardcore, opinion leaders, and media. Iterative update rules of their attribution "opinion", "willingness to express", and "online relationship network" are different. Aiming at the four core research questions, a total of 12 experimental scenarios were designed and simulated by using NetLogo software. Additionally, the simulation model was also thoroughly evaluated for its effectiveness.Regarding the online context: (a). The simulation results from 100 iterations revealed a steady-state characteristic of the number of opinion interactions. That is, whether there is an online context or not, the "spiral of silence" and "polarization" basically no longer repeated after individuals had perceived the opinion climate and media climate for an average of 10 times. (b). In comparison, the "spiral of silence" was more likely to occur when there was no network structure. In the controlled experimental scenario, network context did reduce the "spiral of silence", but at the cost of exacerbating "polarization".Regarding the group composition: (a). Different from previous studies, this study found that the role of opinion leaders is not as great as that of hardcore. Such "resistors of silence" can launch a surprise attack and even reverse the "spiral of silence", while intensifying the "polarization". (b). When simulating the mechanism of "media intervention", the reduction of "polarization" was accompanied by the intensification of the "spiral of silence". And this mechanism had almost no effect on groups that always held neutral opinions. (c). In addition, in the two scenarios where the "opinion" or "willingness to express" of the media was no longer constant, there existed a phenomenon of "double spiral of silence", where the media and other individuals each formed a spiral, resulting in a reciprocal interaction of growth and decline. The excessive "silence" and "reversal" of the media can cause an imbalance in the online interaction.Regarding the network structure: (a). Two types of networks were simulated: a scale-free network with power-law distribution (Barabási-Albert model) and a small-world network with a bell-shaped normal distribution (Watts-Strogatz model). Individuals were more likely to remain silent in the former, typically as isolated nodes, and their opinions generally belonged to the same community. (b). In addition, two types of network update mechanisms were simulated, and the results showed that: Compared with the mechanism of “seeking new potential friends online”, the mechanism of "preferential attachment" and “seeking opinion communities” were the main driving forces behind the "spiral of silence" and "polarization". At the same time, the "echo chamber effect" was also observed.Furthermore, regarding "social media fatigue": the simulation results suggested that too active or too dull interaction groups both can lead to "polarization". On the one hand, when "social media fatigue" is too weak, active online relationship network updates can help suppress the "spiral of silence" and lead to a process of opinion reversal. However, negative opinions would also be easier to spread. On the other hand, when "social media fatigue" is too strong, public opinion would be likely to be dominated by a single opinion, and there may be a separation between online public opinion and real public opinion amidst silence and polarization. A phenomenon called "meta-silence" — staying silent about silence, should be wary of.
