符号网络社区发现算法研究

发布时间：2018-01-09 00:19

本文关键词：符号网络社区发现算法研究　出处：《燕山大学》2016年硕士论文　论文类型：学位论文

【摘要】：真实世界的很多复杂系统都可以用网络来刻画,比如,神经网络、社会网络等都属于复杂网络。最近几年,关于社会网络的社区发现在国内外掀起了一股热潮,社会网络是指社会成员之间因互动等行为而形成的一种相对稳定的关系体系。这些网络通常都具有一个共同的特征社区结构,社区结构表现为将网络划分为大小不均的若干个社区。符号网络的社区结构要保住社区内部主要是正连接,社区之间主要是负连接。本文致力于社会网络中符号网络的社区划分算法研究,提出了适用于符号网络及加权的符号网络社区划分算法。首先,分析了符号网络基本结构特征,对社区划分算法分类分析,并深入地分析了几种经典的社区划分算法,对其有了充分的认识。其次,针对Tushar Sharma等人提出的加权符号网络社区发现算法在第二阶段对未聚类节点再聚类时存在的问题,改进算法对未聚类节点再聚类时没有全面考虑邻居节点的聚类情况的问题,使用模块度来评价划分结果。再次,针对把算法应用于符号网络中当选择不同的节点作为初始节点时,网络最终划分的社区结构不同,即算法存在不稳定性,提出初始点选择算法。当把算法拓展到普通全正网络时,可能生成许多节点数目比较少的社区,因此引入聚类密度和聚类的连接系数提高社区划分的效率,使其在符号网络中的划分结果更合理。最后,通过仿真实验对改进的算法在真实数据集上进行实验验证,通过模块度、错误率和挫败值验证算法的有效性与准确性。
[Abstract]:Many complex systems in the real world can be described by network, for example, neural network, social network and so on belong to complex networks. In recent years, the community on social networks has aroused a wave of fever at home and abroad. Social network is a relatively stable relationship system formed by interaction among social members. These networks usually have a common characteristic community structure. The community structure shows that the network is divided into a number of communities with uneven size. The community structure of symbolic network is mainly positive connection within the community. There are mainly negative connections between communities. In this paper, we focus on the community partition algorithm of symbolic networks in social networks, and propose a symbolic network community partition algorithm which is suitable for symbolic networks and weighted symbol networks. First of all. This paper analyzes the basic structural characteristics of symbolic network, classifies and analyzes the community partition algorithm, and deeply analyzes several classical community partition algorithms, and has a full understanding of them. Secondly. The problems of community discovery algorithm of weighted symbol network proposed by Tushar Sharma in the second phase of the unclustered nodes are discussed. The improved algorithm does not fully consider the clustering of neighbor nodes when the unclustered nodes are reclustered, and the modular degree is used to evaluate the partition results. When different nodes are selected as initial nodes, the community structure of the network is different, that is, the algorithm is unstable. An initial point selection algorithm is proposed. When the algorithm is extended to the normal full positive network, many communities with fewer nodes may be generated, so the clustering density and the connection coefficient of clustering are introduced to improve the efficiency of community division. Finally, the improved algorithm is verified on the real data set by simulation experiments, and the module degree is adopted. Error rate and defeat value verify the validity and accuracy of the algorithm.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：O157.5

【相似文献】