加权网络的链接预测问题研究

发布时间：2018-05-29 07:50

本文选题：复杂网络 + 加权网络　；参考：《南京邮电大学》2017年硕士论文

【摘要】：链接预测是复杂网络分析或链接挖掘问题的一个子问题,也是其中的重点任务之一。链接预测应用范围广泛,得到不同学科研究者的关注,并出现了不同类型的多种方法。其中,基于相似度的链接预测方法适用于大型网络,不仅具有较好的预测准确度,而且具有较低的计算复杂度且较容易实现,现已成为链接预测问题的主要方法之一。已有基于相似度的链接预测方法大多只适用于无权网络,针对加权网络的相关工作很少且预测性能也不理想。很多现实网络的链接带有权值,包含了更多的信息。比如,在线社交网络中,一对好友之间链接的权值代表某段时间内这对好友联系的次数。权值越大,则关系越紧密。在这一背景下,本课题研究加权网络中的链接预测问题,并提出两种改进的加权网络链接预测方法。本课题首先提出一种基于改进加权相似度指标的链接预测方法。已有的基于相似度的加权指标较为简单,仅将权值直接累加。本课题考虑到两个节点与其共同邻居节点间链接权值的差异对这两个节点间形成新链接的贡献并不相同,对加权相似度指标作了改进。实验表明,基于新的相似度加权指标的链接预测方法具有更优的性能。最近,弱关系理论开始应用于加权网络的链接预测问题,并取得了一定的效果。然而,如何在加权网络中区分强弱关系目前还未得到很好的解决。本文提出了一种基于社团检测的加权网络链接预测方法。首先检测出加权网络中的社团,随后根据不同社团之间关系较弱而同一社团内部关系较强这一原理,对链接的强弱进行度量,再使用改进的加权相似度指标进行链接预测。实验表明,新方法的性能要优于未采用弱关系理论的基准方法。
[Abstract]:Link prediction is not only a sub-problem of complex network analysis or link mining, but also one of the key tasks. Link prediction is widely used, and has attracted the attention of researchers in different disciplines, and different types of methods have emerged. Among them, the similarity based link prediction method is suitable for large networks. It not only has better prediction accuracy, but also has lower computational complexity and is easy to implement, so it has become one of the main methods of link prediction problem. Most of the existing link prediction methods based on similarity are only suitable for weighted networks, and the performance of prediction is not satisfactory. Many real-world links have weights and contain more information. In online social networks, for example, the weight of a link between a pair of friends represents the number of times the pair contacted each other over a certain period of time. The greater the weight, the closer the relationship is. In this context, the problem of link prediction in weighted networks is studied, and two improved weighted network link prediction methods are proposed. In this paper, a link prediction method based on the improved weighted similarity index is proposed. The existing weighted index based on similarity is simple and only accumulates the weight directly. Considering that the difference of link weights between two nodes and their common neighbors is not the same contribution to the formation of new links between the two nodes, the weighted similarity index is improved. Experiments show that the link prediction method based on the new similarity weighting index has better performance. Recently, the weak relation theory has been applied to the link prediction problem of weighted networks, and some results have been obtained. However, how to distinguish between strong and weak in weighted networks has not been well solved. This paper presents a weighted network link prediction method based on community detection. First, the associations in the weighted network are detected. Then, according to the principle that the relationship between different communities is weak and the internal relationship of the same community is strong, the strength of the link is measured, and the link prediction is carried out by using the improved weighted similarity index. The experimental results show that the performance of the new method is better than that of the reference method without weak relation theory.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O157.5

【相似文献】