Temporal link prediction algorithm based on local random walk

Abstract

Link \u00a0prediction \u00a0is \u00a0an \u00a0important \u00a0part \u00a0of \u00a0complex \u00a0network \u00a0research. \u00a0Traditional \u00a0static \u00a0link prediction algorithm ignores that nodes and links in network are added and removed over time. But temporal link \u00a0prediction \u00a0can \u00a0use \u00a0the \u00a0information \u00a0of \u00a0historical \u00a0network \u00a0to \u00a0make \u00a0better \u00a0prediction. \u00a0Based \u00a0on \u00a0local random walk, this paper proposes a time-series random walk algorithm. Given link data for times 1 through T, then we predict the links at time T+1. The algorithm first computes the Markov probability transfer matrix at each time, then combines them into a transformation matrix, and applies the local random walk algorithm to obtain \u00a0the \u00a0final \u00a0prediction \u00a0result. \u00a0The \u00a0experimental \u00a0results \u00a0on \u00a0real \u00a0networks \u00a0show \u00a0that \u00a0our \u00a0algorithm demonstrates better than other algorithms.