基于奇异值分解的含权网络匿名化的安全性分析

摘要

分析基于奇异值分解(SVD)的匿名方法在加权社交网络隐私保护中的安全性,给出在含整数权重网络中的重构方法和在含任意权重网络中的非精确重构方法,定义 ε-N -容忍性来衡量其安全性,指出目前谱分析理论得到的ε(可重构系数)上界过于保守因而缺乏指导性.通过实验来测试随机网络、Barabasi-Albert网络、小世界网络以及实际网络的可重构系数,同时测试了基于 SVD 的双重扰动策略的可重构系数.实验结果表明,加权社交网络对谱的丢失具有不同的容忍性,其容忍性与网络参数之间存在密切的关系.%The security of anonymous method based on singular value decomposition (SVD) in the privacy preserving of weighted social network was analyzed. The reconstruction method in network with integer weights and the inexact recon-struction method in network with arbitrary weighted were proposed.The ε-N -tolerance was definited to measure its safety.It was also pointed out that the upper bound of ε (the reconfigurable coefficient) obtained in current spectral theories was so conservative that lacks of guidance. The reconfigurable coefficients of random networks, Barabasi-Albert networks, small world networks and real networks were calculated by experiment. Moreover, the reconfigurable coefficients of double per-turbation strategies based on SVD were also tested. Experimental results show that weighted social networks have different tolerances on spectrum loss, and there is a close relationship between its tolerance and network parameters.