社会网络上连续观点动力学演化及在线预测应用

发布时间：2018-08-31 09:47

【摘要】：近年来,采用自然科学的研究方法研究复杂的社会群体行为和现象引起了人们广泛的兴趣和关注。观点动力学采用基于Agent的建模和仿真方法研究社会群体中个体观点从无序到有序的演化过程。观点是指个体对事物和问题的看法、选择或偏好。观点动力学认为个体观点的形成和演化受到自身和周围其他个体观点的共同影响,试图建立个体观点影响模型,分析和解释观点的传播、演化以及观点簇的涌现。深入研究观点动力学对加深人们对观点形成和演化的内在机理的认识和理解,针对大规模人群设计更好的决策和讨论过程,引导和控制社会舆论的传播和共识的达成等具有重要的理论价值和实践意义。本文研究连续观点动力学演化的理论和应用问题,主要包括以下四个方面的内容:1.有向无标度社会网络上自由发言的观点形成研究。群体中个体的交互关系形成一个有向无标度网络。个体以随机次序或概率次序依次发言表达观点,其他关注发言者的个体接收到提供者的观点,然后依据一定的规则更新自己的观点,在轮到他们发言时进行表达。基于有界信任假设提出一个有向社会网络上的连续观点动力学模型,研究发言次序以及社会网络的拓扑结构对观点形成和演化的影响。结果表明:当同一个个体不断地重复其观点,使得其他关注他并且信任水平在初始时刻包含他的观点的个体们的观点不断地接近他的观点;采用概率次序比随机次序形成更少的观点簇、更大的最大簇、更小的标准差以及只需更少的等待时间达到中等程度的共识;随着无标度分布参数的减少或者信任水平增加,两种发言次序的结果都变好;结果间的差距随着网络规模的减少而变小。2.按轮发言的观点演化研究。针对连续观点动力学模型中缺乏对个体信任及其观点的不同权重的考虑,通过引入个体间的信任度和观点间的相似度,修改有限信任假设为有限影响假设,扩展Hegselmann-Krause(HK)模型为带权重的观点更新模型,以研究社会群体中少数的偏执个体和权威个体对观点形成、演化以及共识构建过程的影响。仿真结果表明,这两类个体的初始观点尽量接近观点分布区间中点0.5或者增加偏执个体的影响阈值和权威个体的可信任程度可以使群体形成更少和更大的观点簇。通过仿真比较按轮发言变式和自由发言变式表明:在相同的初始条件下,自由发言变式采用权威专家优先,然后按观点累积变化最大者优先的发言顺序规则生成的观点簇数更少,15轮左右的发言轮数是观点簇数变化速率的阈值。3.连续观点与有向自适应网络共同演化研究。通过扩展HK模型来研究连续观点和有向自适应Erd?s-Rényi随机网络的共同演化。当2个观点之差超过信任水平?时,模型中的有向链接可能以一定的概率断开。新链接在那些没有有向链接的个体间随机产生。仿真结果表明在静态网络中,最终的观点受到网络的初始特性的影响,而在有向自适应网络中,最终观点基本上受到重连概率的影响。重连概率增加使得网络的平均度增加,加强了个体间的联系,使得自适应网络的结果优于静态网络。4.基于连续观点动力学的在线评分人数预测研究。在线评分中,个体的观点受到自身初始观点和群体观点的共同影响,产生的最终观点将决定个体是否加入评分群体,如果加入将产生评分行为,进而影响后续个体的观点及行为。据此过程建立一个简单的连续观点动力学模型,对在线评分的人员数量进行预测。采用互联网上豆瓣网站的影片在线评分数据进行实验,结果表明模型能够有效预测在线评分人数;个体的最终观点主要受群体差-中-好评分观点的影响,而与自身初始观点基本无关;泊松参数值偏离最优值(1.25)越远,预测准确率越低。
[Abstract]:In recent years, the study of complex social group behaviors and phenomena by natural science has attracted extensive interest and attention. View dynamics uses Agent-based modeling and simulation to study the evolution of individual views from disorder to order in social groups. Opinion Dynamics holds that the formation and evolution of an individual's views are influenced by both his own and other individuals'views around him. It attempts to establish a model to analyze and explain the propagation, evolution and emergence of opinion clusters. Understanding and understanding, designing a better decision-making and discussion process for large-scale population, guiding and controlling the dissemination of public opinion and reaching consensus are of great theoretical and practical significance. Research on the Formation of Free Speaking Views on Social Networks. Interaction among individuals in a group forms a directed scale-free network. Based on the bounded trust hypothesis, a continuous viewpoint dynamics model on directed social networks is proposed to study the effects of speech order and topological structure of social networks on viewpoint formation and evolution. The opinions of the individuals who engrave his opinions are constantly approaching his opinions; using probabilistic order to form fewer opinion clusters, larger maximal clusters, smaller standard deviations, and to achieve a moderate degree of consensus with fewer waiting times than using random order; and with the reduction of scale-free distribution parameters or the increase of trust levels, the two statements The sequential results are better; the gap between the results decreases with the decrease of the network size. 2. Research on the evolution of opinion in turn. In view of the lack of consideration of different weights of individual trust and opinions in the continuous view dynamics model, the finite trust hypothesis is modified by introducing the trust degree between individuals and the similarity between opinions. Limited Impact Hypothesis (LIH) extends the Hegselmann-Krause (HK) model to a weighted view updating model to study the effects of a small number of paranoid and authoritative individuals on opinion formation, evolution and consensus building processes in social groups. Simulation results show that the initial views of the two groups are as close as possible to the midpoint 0.5 or increase in the opinion distribution interval. The influence threshold of paranoid individuals and the trustworthiness of authoritative individuals can make the group form fewer and larger opinion clusters. The simulation results show that under the same initial conditions, authoritative experts are preferred in the round-robin variant and the free-spoken variant, and then the speaker who has the greatest cumulative change of opinion prefers to speak smoothly. The number of viewpoint clusters generated by the order rules is less, and the number of speakers around 15 rounds is the threshold of the change rate of viewpoint clusters. 3. The co-evolution of continuous viewpoints and directed adaptive networks is studied by extending the HK model. The simulation results show that in a static network, the final point of view is affected by the initial characteristics of the network, while in a directed adaptive network, the final point of view is basically affected by the reconnection probability. The increase makes the average degree of the network increase, strengthens the connection between individuals, and makes the result of the adaptive network better than that of the static network. 4. Research on online scoring population prediction based on continuous view dynamics. Whether to join the scoring group or not, if joined, will produce scoring behavior, and then affect the opinions and behaviors of subsequent individuals. A simple continuous view dynamic model is established to predict the number of people scoring online. Individuals'final opinions are mainly influenced by the group's poor-middle-good points, but have little to do with their own initial opinions. The farther the Poisson parameter value deviates from the optimal value (1.25), the lower the prediction accuracy.
【学位授予单位】：国防科学技术大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP393.02

【相似文献】