带协变量的非负矩阵分解的社区发现模型

发布时间：2018-09-19 15:50

【摘要】：至今,我们迈入了复杂网络时代,生活中存在着各种各样的复杂网络,比如移动通信网络、交通网络、电力网络等等。我们的生活和生产活动也越来越多的依赖于这些复杂网络系统安全、可靠、高效的运行。重要的是许多研究表明,这些看上去不尽相同的网络有着共性的概念、理论和方法,这使得研究复杂网络成为必要。网络科学已经成为了一个新的研究领域,并且取得了迅猛的发展。目前来看,网络科学主要研究内容包括发现网络性质、建立网络模型、分析网络行为和设计网络性能。社区结构是许多实际网络具有的一个共同性质,也就是说,整个网络由若干个社区构成,每个社区内部的节点之间连接相对紧密,社区之间的节点连接相对稀疏。因此,有必要研究社区的定量刻画以及复杂网络社区结构的有效挖掘算法。近年来,社区发现技术作为社区结构的挖掘算法,在数据挖掘中是一个极为重要的技术,而且在社会学、生物学、市场学等领域有着广泛的应用,蕴含着巨大的商机。本文会将网络信息和节点特征信息都加入到社区发现模型中。使用基于非负矩阵分解的方法,加入协变量,建立带协变量的非负矩阵分解社区发现模型,进行社区发现。文中研究表明,节点协变量可以灵活的加入到新模型中。也就是,根据不同协变量影响社区结构的重要程度不同,协变量以不同的权重加入到模型中;另一方面,模型可以选择出对社区结构有影响的协变量,淘汰没有影响的协变量。从文中也可以看到,带协变量的非负矩阵分解方法在模拟和实例应用中取得了很好的效果。
[Abstract]:Up to now, we have entered the complex network era, and there are various complex networks in our life, such as mobile communication network, transportation network, power network and so on. Our daily life and production activities depend more and more on the security, reliability and efficiency of these complex network systems. What is important is that many studies show that these seemingly different networks have common concepts, theories and methods, which make it necessary to study complex networks. Network science has become a new research field, and has made rapid development. At present, the main research contents of network science include discovering the nature of network, establishing network model, analyzing network behavior and designing network performance. Community structure is a common property of many real networks, that is, the whole network is composed of several communities, the connections between nodes within each community are relatively close, and the connections between communities are relatively sparse. Therefore, it is necessary to study the quantitative characterization of community and the effective mining algorithm of complex network community structure. In recent years, as a mining algorithm of community structure, community discovery technology is an extremely important technology in data mining, and it has been widely used in sociology, biology, marketing and other fields, which contains huge business opportunities. In this paper, both network information and node feature information are added to the community discovery model. Based on the method of non-negative matrix decomposition, the community discovery model of non-negative matrix factorization with covariable is established, and the community discovery is carried out. The research shows that the node covariable can be added to the new model flexibly. That is, according to the importance of different covariables affecting community structure, covariables are added to the model with different weights; on the other hand, the model can select the covariables that have influence on the community structure and eliminate the unaffected covariables. It can also be seen from the paper that the nonnegative matrix decomposition method with covariables has achieved good results in simulation and practical application.
【学位授予单位】：东北师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：C81

【相似文献】