Monte Carlo methods of PageRank computation
Litvak, Nelli · University of Twente Research Information · 2004
We describe and analyze an on-line Monte Carlo method of PageRank computation. The PageRank is being estimated basing on results of a large number of short independent simulation runs initiated from each page that contains outgoing hyperlinks. The method does not require any storage of the hyperlink matrix and is highly parallelizable. We study confidence intervals, and discover drawbacks of the absolute error criterion and the relative error criterion. Further, we suggest a so-called weighted relative error criterion, which ensures a good accuracy in a relatively small number of simulation runs. Moreover, with the weighted relative error measure, the complexity of the algorithm does not depend on the web structure.