A Scalable Parallel HITS Algorithm for Page Ranking

摘要

The Hypertext Induced Topic Search (HITS) algorithm is a method of ranking authority of information sources in a hyperlinked environment. HITS uses only topological properties of the hyperlinked network to determine rankings. We present an efficient and scalable implementation of the HITS algorithm that uses MPI as an underlying means of communication. We then analyze the performance on a shared memory supercomputer, and use our results to verify the optimal number of processors needed to rank a large number of pages for the link structure of the total University of Southern Mississippi (usm.edu domain) web sites.