【作者】 程军军；

【导师】 刘云；

【作者基本信息】 北京交通大学 ， 通信与信息系统， 2013， 博士

【摘要】 摘要：在Web2.0技术和信息网络技术的共同推动下,社交网络应用蓬勃发展,用户规模与日俱增。在社交网络中,作为自媒体的用户可以随时随地采用多种接入方式参与交互。这种新型、灵活、快捷的交互模式极大地缩短了舆论产生、发酵、扩散的时间；同时,参与主体的高度能动性、自组织性、异质性,社会舆论对网络舆论的非线性映射作用等特点,均使得社交网络中的舆论信息传播演化的复杂性和随机性陡增。而传统的舆论模型和研究方法难以准确描述用户微观交互行为,更难以解释网络舆论传播和演化的宏观过程。鉴于此,本文使用交叉学科的思想和方法,对社交网络信息传播模式、影响力用户拓扑分析与挖掘、话题发现及趋势预测等问题进行了深入研究,力图发现和还原社交网络中信息传播的微观与宏观规律,建立能够刻画这些规律的数学模型,并寻找可以促进或抑制信息传播的相关策略。论文的工作有助于了解社交网络中舆论演化的过程,加深对网络上的复杂群体行为的认识,也为复杂系统的理论研究提供一些探索性的结果。论文的研究工作得到了国家自然科学基金项目(No.61172072、61271308)、北京市自然科学基金项目(No.4112045)和高等学校博士学科点专项科研基金(No.20100009110002)的支持,主要工作和创新点包括以下几个方面：1.研究基于强弱关系理论的社交网络信息传播模型。该模型以传染病动力学经典模型SIR为基础,将节点间信息传播概率视作边连接权重的函数,之后结合强弱关系理论和复杂网络理论建立了用于刻画网络中节点状态随时间演化特性的平均场微分方程组。在真实网络中仿真发现：优先选择弱关系并不能有效的提高信息传播效率,但是去掉弱关系后,无论网络拓扑结构还是信息传播均大受影响；信息传播过程中最大传播节点密度对于节点免疫概率和兴趣衰减概率的变化非常敏感,且拟合了三者之间的函数关系；异质型节点在同样的信息传播机制下会呈现出类似于同质的变化趋势。该模型及其性质有助于进一步深刻认识社交网络中信息传播行为,为进一步研究网络舆论传播提供基础。2.建立基于社会记忆性和优先选择性的社交网络信息传播模型。该模型认为社会加强因子(社会舆论参与度)、人际加强因子(传播链路权重)以及记忆效应因子(个体接触信息次数)均会影响个体对于信息的决策过程,从而影响信息扩散。通过在两个真实网络中的仿真得知：网络拓扑特征会影响信息传播强度和范围；社会舆论参与度越高,信息在网络中成功扩散的几率就越大；在平均度和聚类系数较大的网络中,免疫节点起到信息防火墙的作用,在一定程度上抑制信息扩散；社会舆论参与度越大,个体平均接触信息次数越少。模型较好地还原了社交网络中信息交互的基本特征,为进一步研究网络舆论传播提供理论基础和形式参考。3.分析社交网络节点中心性特征并提出用户影响力度量方法。在实证分析两个真实社交网络中四类节点中心性指标分布以及指标之间的相关性的基础上,提出一种新的用于区分节点影响力的基于边权重的局域型指标(LW指标)。该指标强调节点的影响力由邻居的数量和质量共同决定,并且其时间复杂度远低于紧密度和介数。之后借助SIR信息传播模型验证LW指标的有效性,在两个真实网络中的仿真表明：LW指标在挖掘影响力较大节点方面的性能优于度、紧密度和介数；相比于k核数,LW指标区分粒度更细,实用性更强。该方法可为社交网络广告营销、用户兴趣推荐、网络舆情分析等应用领域提供理论支撑。4.研究社交网络中话题发现和趋势预测方法。设计一套轻型的网络热点话题发现系统,并提出基于BPNN的网络话题发展趋势预测算法。通过对真实网络话题时间序列的预测发现,本模型相比于ARIMA预测模型在还原话题发展趋势以及预测准确度方面均有更优的表现。在此基础上,采用小波降噪、BPNN网络结构优化和学习率自适应调整策略等对模型进行优化,测试表明改进模型的预测性能有了大幅度提升。最后,建立基于BPNN和ARIMA的自适应社交网络话题发展趋势预测模型,提高了预测模型的适用范围。这些方法对网络舆论的监测及预警具有一定的实际应用价值。更多还原

【Abstract】 Driven by the fast development of the Web2.0and information network technology, huge numbers of individuals who have been attracted by various popular applications, are crowding into the social networks. As a self-media, users in social networks can participate in the interactions with other individuals anytime, anywhere and by utilizing any access methods. This new, flexible, fast interactive mode can greatly shorten the evolution time in which public opinion is generated, fermented, and disseminated. At the same time, participants have the highly dynamic, self-organization, and heterogeneity characteristics, and social opinion may influence network opinion with the nonlinear mapping relationship. All of above negative factors make the dissemination and evolution process of network opinion becoming more randomly and complicatedly. However, the traditional models and research methods of public opinion are difficult to accurately describe not only the microscopic interaction behavior between users but also the macroscopic phenomenon of dissemination and evolution. In view of this, we use the interdisciplinary ideas and methods to study information dissemination mode in social networks, influential users mining, topic detection and trends forecasting issues, trying to find and restore the information dissemination in social networks, to establish mathematical models which can characterize these laws, and to find relevant strategies which can promote or inhibit information dissemination. Our work may help to understand the evolution process of public opinion in social networks, to understand the complex group behavior deeply, and also provide some of exploratory theoretical results for the study of complex systems.The work of the dissertation is supported by the National Natural Science Foundation of China (No.61172072,61271308), Beijing Natural Science Foundation (No.4112045), and the Specialized Research Fund for the Doctoral Program of Higher Education of China (No.20100009110002). Main contributions of the dissertation are as follows:1. We propose an epidemic model for information dissemination in social networks based on the theory of tie-strength. The model considers infectious probability as a function of the ties strength, and then establishes the dynamic evolution equations which describe the evolution process of different types of nodes. Moreover, we investigate numerically the behavior of the model on a real scale-free social site with the exponent γ=2.2. We verify that the strength of ties plays a critical role in the rumor diffusion process. Specially, selecting weak ties preferentially cannot make rumor spread faster and wider, but the efficiency of diffusion will be greatly affected after removing them. Another significant finding is that the maximum number of spreaders is very sensitive to the immune probability P and the decay probability v. We show that a smaller μ or v leads to a larger spreading of the rumor, and their relationships can be described as the function ln(max(S))=Av+B, in which the intercept B and the slope A can be fitted perfectly as power-law functions of μ. Our findings may offer some useful insights, helping guide the application in practice and reduce the damage brought by the rumor.2. We establish an information dissemination model based on social memory and tie-strength. The model considers that the social reinforcement factor (social participation of public opinion), the memory effect factor (the number of contacting information for individuals) as well as interpersonal relationship (strength of dissemination ties) will affect the decision-making process of the individual, and then affects the information coverage. Some simulations on two real social sites can prove the following conclusions. Firstly, network topology characteristics will affect the strength and coverage of the information dissemination. Secondly, the higher degree of public opinion participation can lead to greater chance for information spreading in the network. Thirdly, the immune node plays the role of information firewall in the network with higher average degree and clustering coefficient. Finally, the average number of contacting information for individuals will decrease with the increasing participation of the publis opinion. The model can restore the basic characteristics of the information exchange in the social networks, and provide a theoretical basis for further study.3. We study the nodes centricity characteristics and identify influential nodes for spreading dynamics. First, we analyze the distribution of four kinds of centrality indicators and their correlations in two typical social sites. Then, we propose a new centricity indicator based on the theory of ties strength, which is named as local weight indicator (LW). The indicator emphasizes the node influence is jointly determined by the quantity and quality of neighbors, and the complexity of LW is far lower than the closeness and betweenness. Moreover, LW has a finer-grained distinction than k-core indicator. We use SIR model to evaluate the performance of LW indicator in two real social sites. Simulations show that our method can well identify the influencial nodes. The method can provide theoretical support for some applications.4. We study the method of hot topic detection and the algorithm of topics’growth trends prediction in social networks. First we design a lightweight system to detecte hot topics, and then put forward a prediction method based on the BPNN. The results of empirical tests show that our approach is more effective than existing method ARIMA in the aspect of forecasting growth trends. Moreover, we apply wavelet denoising method, BPNN network structure optimization and adaptive learning rate adjustment strategies to improve the above model. Results show that the predicting performance of the optimized model has been improved significantly. Finally, we establish of a self-adaptive forecasting model based on the BPNN and ARIMA in social networks, which can expand the scope of its applications. These methods may help people to master the growth trends through public monitoring and early warning.更多还原