Authors: Kazuma TERAMOTO, Takashi NODERA
Abstract: We now study Lanczos algorithm for computing PageRank vector. This algorithm is based on bi-orthogonalization, which transforms nonsymmetric matrix into tridiagonal matrix to computing PageRank. This generates better approximation of the largest eigenvalue at early stage of iterations. We propose practical scheme of Lanczos biorth- ogonalization algorithm with SVD scheme for computing PageRank. Numerical results show that the proposed algorithm converges faster than existing Arnoldi method in the computation time.
Keywords: eigen value problem Lanczos method PageRank