带权网络的个性化PageRank计算

Computing personalized Page Rank in weighted networks

Page Rank是衡量网络节点重要性的指标之一,个性化Page Rank是普通Page Rank的推广形式.目前关于(个性化)Page Rank的研究主要集中在无权网络,而关于带权网络的研究结果较少.有鉴于此,基于矩阵变换和蒙特卡罗方法,分别给出了在静态和动态带权网络中个性化Page Rank计算方法,并从理论上分析了算法的性能.实验结果显示,两种算法都优于传统的幂迭代算法.

PageRank assigns authority weights to each web page based on the web hyperlink structure,while the personalized PageRank is a generalized version of ordinary PageRank. The computation of personalized PageRank vector in unweighted web is well studied in the past decades,but little is known for the case of weighted webs.In this paper,we analyze the algorithms for PageRank computations in static as well as dynamic weighted networks.The al？gorithms are based on matrix transformation or Monte Carlo methods,and are analyzed theoretically for computation performance.Experiments show that the proposed localized algorithm outperforms power iteration and a referenced Monte Carlo method.