Tables

5.1   Wallclock speedups for A^d Extrapolation, for d 1, 2, 4, 6, 8, and Quadratic Extrapolation.

6.1   Statistics about pages in the STANFORD.EDU dataset whose convergence times are quick (t_i ≤ 15) and pages whose convergence times are long (t_i > 15).

7.1   Example illustrating our terminology using the sample url http://cs.stanford.edu/research/.

7.2   Hyperlink statistics on LARGEWEB for the full graph (Full: 291M nodes, 1.137B links) and for the graph with dangling nodes removed (DNR: 64.7M nodes, 607M links).

7.3   The “closeness” as measured by average (a) absolute error, and (b) KDist distance of the local PageRank vectors _J and the global PageRank segments _J, compared to the closeness of uniform vectors _J and the global PageRank segments for _J the STANFORD/ BERKELEY dataset.

7.4   The local PageRank vector _J for the domain aa.stanford.edu (left) compared to the global PageRank segment _J corresponding to the same pages. The local PageRank vector has a similar ordering to that of the normalized components of the global PageRank vector. The discrepancy in actual ranks is largely due to the fact that the local PageRank vector does not give enough weight to the root node http://aa.stanford.edu.

7.5   Running times for the individual steps of BlockRank for c = 0.85 in achieving a final residual of < 10^-3.

7.6   Wallclock running times for 4 algorithms for computing PageRank with c = 0.85 to a residual of <10^-3.

7.7   Number of iterations needed to converge for standard PageRank and for BlockRank (to a tolerance of 10^-4 for STANFORD/ BERKELEY, and 10^-3 for LARGEWEB).

9.1   Threat models and associated experiments.

10.1 Connection Variables.