在线社会网络的动态社区分析与流行度预测

发布时间：2018-05-27 06:27

本文选题：微博 + 模块度　；参考：《太原理工大学》2014年硕士论文

【摘要】：微博作为一种新兴的在线社会网络,为人们提供了一种新的社交方式和信息传播渠道。微博以其独特的用户交互方式和快速的传播速度,吸引了大量的用户,由此演化出了一个巨大的在线社会网络和信息传播网络。微博网络有其独特的网络拓扑结构,挖掘其潜在的社区结构,发现信息传播的规律,已成为当前计算机信息等多学科的研究热点,受到了广泛关注。探索和掌握微博网络中社区结构动态演化机理和信息传播机制,对于掌握信息传播规律、预测传播流行性、发现网络群体事件、网络舆情预警和控制具有重要意义。本文以微博转发构成的信息传播网络为研究对象,从动态社区结构演化和微博流行度两个方面进行研究。首先,本文改进了基于模块度优化的快速社区发现算法。通过对原算法计算节点顺序的比较,发现节点的扫描顺序会直接影响该算法的时间效率。因此,本文引入“重叠度”的概念来刻画邻居节点间的连接强度；并且对算法的节点计算顺序设计了不同的策略,通过减少节点移动,避免不必要的计算,提高算法的时间效率。通过在人造和真实数据集上的实验表明改进的算法可以有效地提高时间效率。其次,本文利用改进的社区发现算法对微博信息传播网络的动态社区演化进行分析。微博信息传播网络是根据微博转发路径构建的网络,区别于传统的关系网络,其更能真实的反映用户间的交互关系。本文将微博信息传播网络以一个月为间隔,按时间切成网络结构快照,分别进行社区发现来分析社区的演化过程。然后,本文对新浪微博数据集进行了实证分析,对不同微博的流行性进行了研究,提出“流行度”的概念。实证分析不同流行度微博的传播速度和各周期转发特点,发现不同流行度微博的传播机制；并且分析了早期微博转发特征信息与微博最终转发数存在正相关的关系,表明了通过早期的微博转发特征信息可以有效的预测微博最终流行度。最后,本文提出了基于支持向量机(SVM)的微博流行度预测模型,结合微博发布者的用户特征和微博在一小时内的转发特征来作为模型的特征值来预测微博最终的流行度。通过实验对不同流行度和时间段的微博进行预测比较,证明模型能够更加准确的预测微博流行度。
[Abstract]:As a new online social network, Weibo provides a new way of social communication and information dissemination. Weibo attracts a large number of users with its unique user interaction mode and rapid propagation speed, and thus evolves a huge online social network and information dissemination network. Weibo network has its unique network topology, mining its potential community structure, discovering the rules of information dissemination, has become the current computer information and other multidisciplinary research hot spots, has received extensive attention. It is of great significance to explore and master the dynamic evolution mechanism of community structure and the mechanism of information dissemination in Weibo network, which is of great significance for mastering the law of information dissemination, predicting the popularity of communication, discovering network group events, and early warning and control of network public opinion. In this paper, the information dissemination network based on Weibo forwarding is studied from two aspects: dynamic community structure evolution and Weibo prevalence. Firstly, this paper improves the fast community discovery algorithm based on modularity optimization. By comparing the order of nodes calculated by the original algorithm, it is found that the scanning order of the nodes will directly affect the time efficiency of the algorithm. Therefore, this paper introduces the concept of "overlap degree" to describe the connection strength between neighbor nodes, and designs different strategies for the node calculation sequence of the algorithm, which can avoid unnecessary calculation by reducing node movement. Improve the time efficiency of the algorithm. Experiments on artificial and real data sets show that the improved algorithm can effectively improve the time efficiency. Secondly, the improved community discovery algorithm is used to analyze the dynamic community evolution of Weibo information dissemination network. The Weibo information dissemination network is constructed according to the Weibo forwarding path, which is different from the traditional relational network, and it can reflect the interaction between users more truthfully. In this paper, the Weibo information dissemination network is cut into a snapshot of the network structure in a one-month interval, and the community discovery is carried out to analyze the evolution process of the community. Then, this paper makes an empirical analysis of Sina Weibo data set, studies the epidemic of different Weibo, and puts forward the concept of "popularity". This paper empirically analyzes the propagation speed and the characteristics of different cycles of Weibo, finds out the transmission mechanism of Weibo with different prevalence, and analyzes the positive correlation between the forwarding feature information of early Weibo and the final forwarding number of Weibo. The results show that early Weibo forwarding feature information can effectively predict the final popularity of Weibo. Finally, this paper presents a prediction model of Weibo popularity based on support vector machine (SVM), which combines the user characteristics of Weibo publishers and the forwarding features of Weibo within an hour as the eigenvalues of the model to predict the final popularity of Weibo. The prediction of Weibo in different prevalence and time periods by experiments shows that the model can predict Weibo prevalence more accurately.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.092

【参考文献】