An Extrapolation Iteration and Its Lumped Type Iteration for Computing PageRank

Recently, computing Google’s PageRank iteratively has been receiving a lot of attention. Many numerical techniques are proposed to remedy the slow convergence of the traditional powermethod. Among them, lumped methods proceed with amuch smaller matrix if the number of dangling nodes is large. In this paper, to further improve the extrapolation process based on matrix trace and the lumped methods, we first propose a newextrapolation approach for computing PageRank and then develop a lumped version to accelerate PageRank computation. Some numerical results of the newapproach validate the theoretical analysis.