A modified power method for the PageRank problem

The PageRank algorithm plays a very important role in modern search engine technology,and it makes use of the power method to compute the principal eigenvector of the Google matrix representing the weblink graph.However,when the largest eigenvalue cannot be well separated from the second one,the power method may perform poorly.This happens when the damping factor is sufficiently close to 1.Therefore,it is worth developing new techniques that are more sophisticated than the power method.In this paper,we propose an improved version of the power method for computing PageRank.Numerical experiments illustrate the efficiency and convergence behavior of the new algorithm.