摘要
为从理论上比较各种预测方法的优劣,分析多个网络演化过程中形成链接的两个节点之间的拓扑距离分布,阐明了传统基于共同邻居相似性指标可有效进行链路预测的机理,从理论上分析了9种基于共同邻居相似性算法的预测上限(可预测性)。通过分析一阶邻居预测算法的局限性和影响链路可预测性的因素,提出了两种基于高阶路径信息的链路预测算法并计算了他们的可预测性指标。从理论上提出了链路的可预测性指标,也通过对实际网络的预测证明了所提链路预测算法的有效性。
Link predictability in complex networks refers to the upper limit for link prediction ac- curacy by using prediction algorithm, and the analysis of link predictability is conductive to com- pare the pros and cons for various prediction methods in theory. In this paper, we analyzed the topological distance distribution between two nodes forming a link during the process of multiple networks evolution, and then illustrated the mechanism for making effective link prediction based on common neighbor similarity index. At last, we analyzed the prediction upper limit (predicta- bility) for nine algorithms based on common neighbor similarity in theory. By analyzing the limi- tation of first order neighbor prediction algorithm and the factor affecting link predictability, we proposed two types of link prediction algorithms based on high order path information and calcu- lated their predictability index. This study proposed the link predictability index in theory, and proved the validity of the proposed link prediction algorithm by predicting real networks.
出处
《复杂系统与复杂性科学》
EI
CSCD
北大核心
2014年第1期41-47,共7页
Complex Systems and Complexity Science
基金
国家自然科学基金(61004104
61104143)
中央高校基本科研业务费专项基金项目(DC120101132
DC13010215)
关键词
复杂网络
链路预测
相似性指标
局域结构
complex networks
link prediction
similarity index
local structure