【作者】 王秀芳；

【导师】 张晓琴；

【作者基本信息】 山西大学 ， 统计学， 2018， 硕士

【摘要】 近年来,人们越发关注网络中的链路预测问题。从虚拟的世界中,通过链路预测的方式,预测真实生活中人们的关系以及行为,这是一个比较有意义的研究方向,研究学者从不同的角度设计不同的指标不断地进行探索,由于描述网络节点间的关系程度的特征非常多,特征选取的不同会影响相似性指标的预测效果,所以相似性的度量指标有很大的发展空间。本文分别在简单无向网络和符号网络这两种类型网络的基础上作了进一步的工作,研究成果如下:(1)在简单无向网络中,由于共同邻居处于待预测节点间的核心位置,所以它的影响力是显而易见的。本文主要从两方面来考虑。一方面,以共同邻居节点的视角展开分析网络的连通情况,由于网络连接越稠密,信息越流通,描述这种现象,本文分析的是两节点的共同邻居节点与两节点的邻居节点之间的连接情况,连边数越多,则信息传递的越多。另一方面,从一个节点到另一个节点的资源分配进行分析,本文假设这种分配是不均匀的,认为两节点的邻居个数越多,向共同邻居索取的资源也就越多。结合这两种思想,提出本文的新指标:BCNI指标。用AUC评价指标进行了实验验证,实际数据分析的结果显示,该方法有效的改进了基于节点局部信息相似性指标。(2)符号网络,即网络中有多种不同类型的边,最典型的是存在正边和负边,怎么预测正边和负边逐渐成为一个越来越重要的研究主题,了解到在社会网络中局部路径指标(LP)表现良好。且结构平衡理论作为符号网络研究的理论基础有一定的现实意义,本文把LP算法和结构平衡理论进行结合应用到符号网络中,用AUC评价指标进行了实验验证,LP指标同样适用于符号网络,并且给出它的平衡理论解释。无论是简单无向网络还是符号网络,把本文所提出的方法应用到真实网络和人工选取的部分真实网络中,经过实验验证都是有效可行的,具有一定的应用价值。更多还原

【Abstract】 In recent years,more and more attention has been paid to the link prediction problem in the network.It is a meaningful research direction to predict people’s relationship and behavior in real life through link prediction in the virtual world.Researchers from different angles design different indicators to explore continuously.Because there are many characteristics of the relationship between network nodes,the difference of feature selection will affect the prediction effect of similarity index,so the measurement index of similarity has a great development space.In this paper,we have done further work on the basis of two types of networks,simple undirected networks and symbolic networks,and the results are as follows:(1)In a simple and undirected network,because the common neighbors are at the core of the nodes to be predicted,its influence is obvious.This article is specifically considered from two aspects.On the one hand,we describe the phenomenon from the connectivity of the common neighbors in the network.In this paper,we take the ratio of the number of common neighbor nodes of two nodes to the number of neighbor nodes of the two nodes.The larger the ratio is,the more information is transmitted.On the other hand,the resource allocation from one node to another is analyzed.This paper assumes that the allocation is not uniform.It is considered that the larger the number of common neighbors of the two nodes,the more resources will be demanded from the common neighbors.Combined with these two ideas,the new index of this article is put forward: BCNI index.The experimental verification is carried out with the AUC evaluation index.The results of actual data analysis show that the method improves the similarity index of local information based on nodes effectively.(2)Symbolic network,that is,there are many different types of edges in the network.The most typical is the positive side and the negative side.How to predict the positive edges and the negative edges has gradually become an increasingly important research topic.Understand that in the social network of local path index(LP)of good performance,and structural balance theory as the theoretical basis for the study of symbolic network has certain practical significance.In this paper,LP algorithm and structural balance theory are applied to the symbolic network.The AUC evaluation index is used to verify the experiment.LP index is also suitable for symbol network,and its balance theory explanation is given.Whether it is a simple undirected network or a symbolic network,the proposed method is applied to real networks and partly selected real networks,which is effective and feasible after experimental verification,and has certain application value.更多还原