【作者】 王茹；

【导师】 蔡皖东；

【作者基本信息】 西北工业大学 ， 计算机科学与技术， 2017， 博士

【摘要】 互联网作为网络信息技术的代表,在过去的几十年中得到了迅速的发展,人类社会已经进入了复杂网络时代。微博网络作为一种复杂网络极大的改变了人类交流和获取信息的方式,人类在享受便利的同时也为网络的安全可靠和有效性带来了挑战。随着微博使用人数的增多,需要解决的安全问题也越来越多。此外,微博网络具有网络规模大,连接复杂,内容多元以及动态变化等特性。由于这些特性的存在就无法沿用传统的基于图论的社团发现算法,因此对社团结构的发现及相关问题研究得到了诸多领域研究者的关注。但是发现微博网络中社团首先要研究微博的信息传播特性和用户行为特性,只有分析了微博的特性才能结合传统的社团发现方法提出一种基于微博特性的社团发现方法,得到的结果也会更加准确和符合实际情况,甚至能够跳出网络结构的束缚,发现网络中潜在的社团。微博网络具有动态变化的特性,即网络结构和用户行为都会随着时间发生变化,这就导致社团结构也在不断的随着时间变化。因此,分析微博社团的动态演化过程,并且对其进行进一步的预测同样具有重要的研究价值和意义。总之,如何从微博网络中发现社团并对社团动态演化过程进行分析和预测成为了本文的重点。具体地,本文的主要研究工作及贡献如下:1、研究了微博网络中信息的传播规律,提出了一种基于传播动力学的微博信息传播模型。该模型通过对复杂网络中SEIR动力学模型进行改进并且重新定义节点状态,不仅实现了对微博网络中信息传播规律的刻画,而且还对传播达到稳定状态进行了量化分析。通过与真实网络进行对比实验表明,该模型描述的微博网络中信息的传播规律与真实情况基本一致。2、研究了微博用户的转发行为特征,提出一种用于分析用户转发行为的博弈模型。该模型是对序贯博弈中继承模型进行改进,通过马尔科夫完美策略——MPE(Markov perfect equilibrium strategy)估算出用户转发信息的概率,不仅实现了对用户转发行为的量化分析,而且还找出了影响转发行为的关键因素。通过在真实网络中进行实验表明,该模型能够描述用户的转发行为,验证了模型的真实性。3、研究了微博网络中社团的结构和特性,提出了一种基于活跃度特性的社团发现方法。该方法首先对用户间转发关系进行极大似然估计,进而提取活跃度特征集合。然后将活跃度特征集合与LOC(Link Optimization Community)算法进行融合,解决了社团成员划分不合理的问题,提高了社团发现的准确性和合理性。该方法不仅能够准确发现社团结构,而且对于微博中隐藏社团的发现也有较好的效果。通过在标准测试网络和真实网络上的实验,验证了该方法的准确性和合理性。4、研究了微博网络中社团用户之间的合作行为对社团的影响,提出了一种基于用户合作行为的社团动态演化分析方法。该方法首先构建一个基于合作行为的社团演化模型,由于社团中边的连接受到用户合作行为的影响而产生变化,因此社团结构在每个时刻都会产生变化。在每一次演化过程中,用户都会对行为策略进行调整,通过分析社会困境收益矩阵构建用户关系和策略调整规则,进而计算出下一时刻的用户关系和策略,实现对社团演化规律的预测,最后通过仿真实验验证了方法的可靠性。最后,在全文研究的基础上,进一步讨论了未来研究的方向。更多还原

【Abstract】 In the past few decades,Internet as a representative of network information technology has been rapid development.The human society has entered a complex network era.As a complex network Microblog has greatly changed the way of human communication and getting information.The humans are enjoying the convenience of the network.At the same time there is the challenge in the safety and reliability of the network.With the increase in the number of using microblog,more and more security issues need to solve.In addition,the large scale,complex connection,multi-content and dynamic change are the characteristics of the microblog network.Because of the existence of these characteristics,the traditional graph-based community detection algorithms are not fully accepted.So the related issues of community detection have been studied in many fields.In order to detect community in microblog,the user behavior characteristics and information dissemination characteristics must be mainly studied.Combining microblog characteristics with the traditional community detection algorithm propose a new community detection method.The results will be more accurate,and the potential communities will be detected.Because network structure and user behavior of microblog will change with time,the community structure will be changed.Therefore,it has important significance and value to analyze and predict the dynamic evolution process of microblog community.Overall,how to detect communities and predict the dynamic evolution process of the community are the research focus in this paper.Specifically,the main researches and contributions of this paper are as follows:1.In this paper,we study the propagation law of microblog information,and propose the microblog information dissemination model based on propagation dynamics.The model improves the SEIR dynamics model in the complex network and redefines the node state.It not only achieve the description of information dissemination,but also the quantitative analysis of spreading stability.Compared with the real network,the experiment shows that the propagation rules of the model are the same as the real situation.2.In this paper,we study the retweeting behavior of microblog users,and propose a retweeting game model.The model improves the inheritance model in sequential game.Using the Markov perfect equilibrium strategy(MPE)estimates the probability of retweeting.It not only achieve quantitative analysis,but also find out the key factors that affect the retweeting behavior.Experiments show that the model can describe the retweeting behavior and verity the authenticity of the model.3.In this paper,we study the structure and characteristics of the community,and propose a community detection method based on activity characteristics.The method uses the large likelihood estimation to extract the retweeting active characteristic set.Then we use the combination of the active characteristic set and Link Optimization Community(LOC)algorithm to solve the problem of unreasonable division.The method improves the accuracy and rationality of the community detection.Therefore,using the proposed method could accurately find the community structure and detect the hidden communities.The accuracy and rationality of the method are verified by experiments on standard test network and real microblog network.4.In this paper,we study the influence of the cooperative behavior among the users in microblog community and propose a dynamic evolution model based on the cooperative behavior.This method constructs a community evolution model based on cooperative behavior.The connection in the community is changed by the influence of user cooperation,which cause the community structure change at every moment.In each evolution process,the user will adjust the strategy.The user builds user relationship and strategy adjustment rules by analyzing the social dilemma income matrix.The user calculates the next moment of user relations and strategies to predict the community evolution.The reliability of the method is verified by simulation experiments.Finally,on the basis of the paper the direction of future research is further discussed.更多还原