【作者】 孟青；

【导师】 刘波； 李国锋；

【作者基本信息】 东南大学 ， 工程硕士（专业学位）， 2017， 硕士

【摘要】 在线社交网络的飞速发展,正在改变着人们的社交方式。然而,由于社交网络的开放性、匿名性,用户发布信息的自由性、趋同性等特点,使得社交网络成为舆论信息的滋生平台。而支撑舆论快速扩散传播的正是由用户好奇心、从众心等特征促成的网络群体。恶意舆论、谣言的传播离不开网络群体的推波助澜,群体影响力开始引起人们的注意,研究者们从群体角度对在线社交网络中的信息传播、影响最大化和群体现象等方面开展了研究,然而还未出现针对群体影响力系统、全面的分析和研究,也没有针对性的建模方法。本文从衡量群体影响力的角度出发,提出一种基于因子图的群体影响力模型,分别从群体和个体的角度量化相关特征,结合在线社交网络中的真实数据,设计了群体影响力模型,以全面、精确地描述群体和个体在在线社交网络中的影响力作用。论文的具体工作包括:首先,建立双层网络模型并进行群体划分。通过设计合理的数据采集框架,采集并整理在线社交网络数据,选取有效的数据集,对在线社交网络形成的拓扑关系进行分析,建立双层网络关系模型(Double-layer Network Relation Model,DNRM),并在此基础上基于Louvain算法设计群体划分算法,分别挖掘各层次中潜在的群体结构,得到个体所属的两类群体集合:关注型群体集合和交互型群体集合。其次,提出基于因子图的群体影响力模型并应用于个体行为预测。通过衡量不同类型群体的属性,结合个体的群体属性,建立群体影响力模型。该模型从群体层次和个体层次全面考虑了群体影响力的作用效果,并把对群体的影响力衡量应用在个体行为预测方面,提出基于群体影响力的个体行为预测算法(Group Influence based Individual Prediction Algorithm,GIBP)。最后,使用在线社交网络真实数据集,把群体影响力的建模方法应用在个体行为预测中并进行实验。通过实验结果得出以下结论:本文提出的群体影响力的衡量方法在个体预测方面能更加精准地预测个体行为趋势,并能够发现更多具有明显群体行为的用户;基于群体影响力的个体行为预测算法与现有的经典算法比较具有更好的预测结果,证明了群体影响力建模方法的合理性和有效性。对群体影响力的研究,不但有利于理解网络群体现象和群体行为,而且有利于分析群体中个体行为变化规律。研究群体对群体内个体的影响作用,也有助于理解复杂社交网络的演变过程。准确地分析和有效地利用群体与个体之间的行为因果关系,能够为市场营销、网络舆论、网络暴力的控制等决策问题提供可靠的理论依据。更多还原

【Abstract】 People’s forms of socializing changed by the rapid development of Online Social Network.However,due to the openness,anonymity,freedom of information releasing and convergence of the social network,it is becoming a platform for public opinion information.It is groups facilitated by curiosity or conformity that lead public opinion spreading quickly.Public malicious opinion and rumors are inseparable fueled by the groups in social network,and group influence has attracted researcher attention in recent years.In the perspective of group,some researchers have carried on the research of the information dissemination,the influence maximization and the group phenomenon in the online social network.However,there is no systematic modeling method for group influence.In terms of group and individual,we quantified features and proposed a group influence model based on factor graph.Combining with the Sina Weibo datasets,a group influence model designed to describe the influence of individual in the online social network.The main work includes:First,a double-layer network relation model is proposed.With data crawling framework fine designed,we collecting and organizing online social network data and selecting effective dataset for experiment.Double-layer network relation model proposed for analyzing the topological relations of online social network,and mining the potential group structure based on the Louvain algorithm.At last,we get two kinds groups:the following groups and the interaction groups.Secondly,we proposed a group influence model and applied to individual prediction.Modeling the group influence based on factor graph by measured the attributes of different types of groups and individual on group level.The model considers the effect of group influence from the group level and the individual level comprehensively,and then applies the group influence method to the individual behavior prediction,with a group influence based individual behavior prediction algorithm,GIBP proposing.Finally,experiments carried out based on GIBP using the online social network dataset.After experiments results analyzed,we get the following conclusions:The group influence model proposed in this paper can depict the influence of users more effectively and more accurately,and can find more users’ with obvious group behavior.The GIBP algorithm has better experimental results compared with the existing classical algorithms,which proves the rationality and validity of the group influence modeling method proposed in this paper.The research of group influence does not only help to understand group phenomenon and behavior in social network,but also helps to analyze the individual behavior trends in groups.It is more helpful to understand the evolution process of the complex social network,and it will provide reliable for the decision-making problems such as marketing,network public opinion and cyber violence control with making effective use of the causal relationship between the group and the individual.更多还原