Parallel adaptive technique for computing PageRank

Re-ranking the search results using PageRank is a well-known technique used in modern search engines. Running an iterative algorithm like PageRank on a large Web graph consumes both much computing resource and time. This paper therefore proposes a parallel adaptive technique for computing PageRank using the PC cluster. Following the study of the Stanford WebBase group on convergence patterns of PageRank scores of pages using the conventional PageRank algorithm, PageRank scores of most pages converge more quickly than the remainder, we then devise our parallel adaptive algorithm to reiterate the computation for pages whose PageRank scores are still not converged. From experiments using a synthesized Web graph of 28 million pages and around 227 million hyperlinks, we obtain the acceleration rate up to 6-8 times using 32 PC processors.