Efficient Parallel PageRank Algorithm for Network Analysis

We propose an efficient version of the PageRank algorithm for adjacency matrices, that reduces the complexity by a factor two. This method computes the
$A^{T}x$
operation on the transpose matrix
$A^{T}$
without having to explicitly normalize and transpose the matrix. We implement the method using standard row-major and column-major matrix storage formats. We perform experiments with parallel implementations in OpenMP, on synthetic data as well as on matrices extracted from large-scale graphs. The experiments are done on two different Intel processors from recent generations. The columnmajor storage format version of our method shows good scaling and outperforms the standard PageRank in a majority of cases, even when not considering the preprocessing burden in the latter.