Link prediction from partial observation in scale-free networks

Abstract

We study the link prediction problem in scale-free networks by using node similarity measure method to estimate the probabilities of potential links from a partial observation of links. Specifically, we give estimates of scaling parameter of the power-law distribution based on the observed node degrees and approximate solutions to a transcendental equation with Hurwitz-zeta function, whereby to obtain a local sub-network similarity measure of node pairs that do not have available observation information. Experiments on synthetic scale-free networks verify the effectiveness of our method.