一种基于加权网络拓扑权重的链路预测方法

近年来,复杂网络中的链路预测问题受到越来越多的关注,链路预测的应用场景也越来越广泛,因此如何提高链路预测精度是一个重要问题。目前已提出了很多方法,其中加权相似性指标的预测方法取得了很好的效果。然而传统的加权网络链路预测方法仅考虑了链接的自然权重,忽略了链接的拓扑权重对预测精度的影响。因此,针对加权网络的链路预测,综合考虑网络中边的聚类和扩散特性并将其作为边的拓扑权重,提出了基于链接拓扑权重的WCD含权预测指标,包括WCD-CN,WCD-AA,WCD-RA和WCD-LP4个相似性指标。文中以Matlab为实验平台,在两个带权数据集(USAir,Bibble)和两个无权数据集(Pblogs,Dolphins)上进行实验,并以AUC作为评价指标。仿真结果表明,与基于自然权重的含权指标、基于簇系数的结构含权指标相比,所提算法具有更好的预测精度。

Link Prediction Method Based on Weighted Network Topology Weight

In recent years,with more and more attention drawning to link prediction in complex networks,and with the application of link prediction becoming increasingly extensive,a crucial question is raised on how to improve the accuracy of link prediction.Many proposals are made,among which the weighted similarity indices have already achieved a promising result.However,the traditional weighted network link prediction only considers the natural weight of the link neglects the influence of the topologi-cal weights on prediction accuracy.Therefore,aiming at the weighted networks,this paper takes the clustering and diffusion characteristics of edges into consideration and regard them as the topological weights of edges,and consequently recommended four similarity indices based on the topology weight of links,namely WCD-CN,WCD-AA,WCD-RA,and WCD-LP.This paper takes Matlab as the experimental platform and carries out experiments on two weighted datasets(USAir,Bibble)and two weightless datasets(Pblogs and Dolphins),in which AUC is used as the evaluation index.The results of the simulation indicate that compared with two weighted indices,which are based on natural weight and cluster coefficient respectively,the proposed algorithm has higher accuracy in prediction.