基于置信传播和依赖协变量的网络社区发现方法

发布时间：2018-07-12 20:53

本文选题：网络社区发现 + 随机分块模型　；参考：《东北师范大学》2017年硕士论文

【摘要】：网络在我们的生活中随处可见,很多系统都可以用网络来刻画,比如社会网络,生物网络,通信网络等等。随着以互联网为代表的网络信息技术的迅速发展,人类的生活与生产活动越来越多地依赖于各种网络。我们急需深入地挖掘网络,对网络数据进行深度分析来解决现实网络问题。而网络社区发现就是网络数据挖掘的一个重要课题,网络社区发现方法就是把网络中相似节点分为一个集合,称之为社区,使得社区内节点间的连接更为稠密,而不在一个社区的节点之间的连接极为稀疏。我们试图通过网络的连接情况以及各种节点的信息来发现这个网络里的社区分类情况,网络社区发现的成功将会使网络更加明晰也更加富有解释性,同时其社区发现会给现实网络带来很多的后续开发价值,比如精准营销等等。随着大数据时代的蓬勃发展,人们对于网络社区发现方法的要求越来越高,许多相关学者都致力于研究更为高效更为精确更有实际价值的网络社区发现方法。本文在传统的随机分块模型中引入了协变量信息,也就是说要通过网络和网络中节点的协变量共同来发现社区,这也更符合现实事实与实际需求。本文通过对随机分块模型社区发现的两种方法的等价性证明,来肯定了置信度传播思想在网络社区发现里的可行性,于是对于这样一个依赖于协变量的网络社区发现模型,在大型稀疏网络环境里,我们采用置信传播方法来完成参数估计。由于置信传播方法在带环的网络结构中并没有合理的解释性,所以鲜少被应用于网络社区发现领域,然而对于大型稀疏网络,我们更关心的是在复杂度低、操作简易的前提下的模型精确度,通过文章的生成模型实验以及基金委统计学关键词网络的实验结果我们也可以看到,推广到所建模型的置信传播算法比传统的EM算法还有谱方法等更为迅速,结果也同样极为精确。
[Abstract]:Network can be seen everywhere in our life, many systems can be described by network, such as social network, biological network, communication network and so on. With the rapid development of network information technology represented by the Internet, human life and production activities depend more and more on various networks. We need to dig into the network and analyze the network data to solve the real network problem. The discovery of network community is an important subject of network data mining. The method of network community discovery is to divide the similar nodes in the network into a set, which is called community, which makes the connection between nodes in the community more dense. Connections between nodes not in a community are extremely sparse. We try to find out the classification of the communities in the network through the connection of the network and the information of the various nodes. The success of the discovery of the network will make the network clearer and more explanatory. At the same time, its community discovery will bring a lot of future development value to the real network, such as precision marketing and so on. With the rapid development of the big data era, people are demanding more and more online community discovery methods. Many related scholars are devoted to the research of more efficient, more accurate and more practical online community discovery methods. In this paper, covariable information is introduced into the traditional random block model, that is to say, it is necessary to discover the community through the covariables of the network and the nodes in the network, which is also more in line with the reality and the actual demand. In this paper, we prove the equivalence of the two methods of community discovery in random block model, and confirm the feasibility of confidence propagation in the discovery of network community, so for such a covariable-dependent network community discovery model, In the large sparse network environment, we use confidence propagation method to complete the parameter estimation. Because the confidence propagation method has no reasonable explanation in the ring network structure, it is rarely used in the field of network community discovery. However, for large sparse networks, we are more concerned about the low complexity. The accuracy of the model under the premise of simple operation, we can also see the results of the model experiment and the experimental results of the CSC statistical keyword network. The confidence propagation algorithm extended to the established model is more rapid than the traditional EM algorithm and spectral method, and the results are also extremely accurate.
【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】