To randomize or not to randomize: space optimal summaries for hyperlink analysis