基于相似性的链路预测方法及应用

发布时间：2018-06-26 06:13

本文选题：链路预测 + 复杂网络　；参考：《中国计量大学》2016年硕士论文

【摘要】：本论文研究基于节点相似性的链路预测(Link Prediction),网络中的链路预测既包含对未知链接(记录网络过程中遗漏或被隐藏的边)的预测,也包含对未来链接(未来很有可能存在的边)的预测.针对链路预测的方法及应用,我们基于节点对相似性提出了两个新的局部相似性指标;随后考虑链路预测对重要连边的预测功能,我们将链路预测可指导重要连边的思想运用到崩溃网络的恢复过程中.受网络社团结构的启发,本文首先提出一个新的局部相似性指标——局部社团结构(Local Community Structure)指标,简称为LCS指标.在已知网络局部信息的前提下,LCS指标刻画了网络中任意两个节点间的共同邻居节点与这两个节点的聚集关系.在基于真实网络的实验中,我们计算并比较了CN、AA、RA和LCS四个指标在7个不同真实网络中的AUC值.我们发现,在簇系数较大的网络中,LCS指标的预测精确度要优于其他三个指标.在上述LCS指标的构建基础上,本文借助节点的邻居节点与另一个节点之间的紧密程度来定义两个节点间的相似性,提出了CS指标.该指标结合以往指标的思想和技巧,通过数据实验显示,它在大多数真实网络中可达到比其他指标更高的预测精确度.链路预测算法不仅可以用于预测网络中缺失的连边,也可以用于探索网络中隐藏的链接以及重要的链接.对网络中重要链路的探索研究,本文主要以电力网络的实际背景为主,在大规模瘫痪状态下的电力系统的恢复过程中,网络中的一些特殊连边起到了关键作用,这也是本文提出的基于异常链路分析的网络重构策略的主要思想.通过链路预测算法对网络中真实存在的连边进行异常度排名,以优先恢复异常度高的电源节点为目标,建立骨架网络恢复策略,然后根据链路的重要性进行骨架网络之外的线路的修复.这样不仅可以快速连通电源发电机,也能及时恢复重要线路,具有实际意义.
[Abstract]:In this paper, link prediction based on node similarity is studied. The link prediction in the network includes the prediction of unknown links (recording the missing or hidden edges in the network). It also includes predictions of future links (which are likely to exist in the future). Aiming at the method and application of link prediction, we propose two new local similarity indexes based on node pair similarity, and then consider the prediction function of link prediction to important connected edges. We apply the idea of link prediction to the restoration of crash networks. Inspired by the network community structure, a new local similarity index, Local Community structure (LCS), is proposed in this paper. Based on the known local information of the network, the LCS index describes the aggregation relationship between the common neighbor nodes and the two nodes between any two nodes in the network. In the experiment based on real network, we calculate and compare the AUC values of CNA AARA and LCS in 7 different real networks. We find that the prediction accuracy of LCS is better than the other three indexes in the network with large cluster coefficient. Based on the above LCS index, the similarity between the two nodes is defined by the tightness between the neighbor node and the other node, and the CS index is proposed. Combined with the ideas and techniques of the previous indexes, the data experiments show that the index can achieve higher prediction accuracy than other indexes in most real networks. The link prediction algorithm can be used not only to predict the missing links in the network, but also to explore the hidden links and important links in the network. For the exploration and research of important links in the network, this paper mainly focuses on the actual background of the power network, and plays a key role in the restoration process of the power system under the large-scale paralysis. This is also the main idea of the network reconfiguration strategy based on abnormal link analysis. By using the link prediction algorithm to rank the abnormal degree of the real connected edges in the network, the restoration strategy of skeleton network is established with priority to recover the power node with high anomaly degree. Then, according to the importance of the link, the repair of the line outside the skeleton network is carried out. It is of practical significance not only to connect the power generator quickly, but also to restore the important circuit in time.
【学位授予单位】：中国计量大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：O157.5

【相似文献】