面向结构洞的复杂网络关键节点排序  

Ranking key nodes in complex networks by considering structural holes

作　　者：韩忠明[1] 吴杨[1] 谭旭升 段大高[1] 杨伟杰[1] 

机构地区：[1]北京工商大学计算机与信息工程学院,北京100048

出　　处：《物理学报》2015年第5期421-429,共9页Acta Physica Sinica

基　　金：国家自然科学基金(批准号:61170112);中央财政支持地方高校发展专项资金人才培养和创新团队建设项目(批准号:19005323132);教育部人文社会科学研究基金项目(批准号:13YJC860006)资助的课题~~

摘　　要：复杂网络中的结构洞节点对于信息传播具有重要作用,现有关键节点排序方法多数没有兼顾结构洞节点和其他类型的关键节点进行排序.本文根据结构洞理论与关键节点排序相关研究选取了网络约束系数、介数中心性、等级度、效率、网络规模、Page Rank值以及聚类系数7个度量指标,将基于List Net的排序学习方法引入到复杂网络的关键节点排序问题中,融合7个度量指标,构建了一个能够综合评价面向结构洞节点的关键节点排序方法.采用模拟网络和实际复杂网络进行了大量实验,人工标准试验结果表明本文排序方法能够综合考虑结构洞节点和核心节点,关键节点排序与人工排序结果具有较高的一致性.SIR传播模型评估实验结果表明由本文选择TOP-K节点发起的传播能够在较短的传播时间内达到最大的传播范围.Structural hole nodes in complex networks play important roles in the network information diffusion. Unfortunately, most of the existing methods of ranking key nodes do not integrate structural hole nodes and other key nodes. According to the relevant research on structural hole theory as well as the key node ranking methods, network constraint coefficient, betweenness centrality, hierarchy, efficiently, network size, PageRank and clustering coefficient, 7 metrics are selected to rank the key nodes. Based on the 7 metrics, a ranking learning method based on ListNet is introduced to solve ranking key nodes by multi metrics. Comprehensive experiments are conducted based on different artificial networks and real complex networks. Experimental results with manual annotation show that the ranking method can comprehensively consider the structural hole nodes and other nodes with different important features. The ranking results on different networks are highly consistent with the manual ranking results. The spreading experiment results using signed to interference ratio propagation model show that SIR model can reach a maximum propagating ratio in a shorter propagating time initiated by TOP-K key nodes selected by our method than TOP-K key nodes selected by other methods.

关 键 词：结构洞 关键节点 排序学习 复杂网络 

分 类 号：O157.5[理学—数学] O223[理学—基础数学]