The document proposes methods to accelerate PageRank computations through extrapolation. It discusses using successive PageRank iterates to estimate the components of the current iterate in the directions of the eigenvectors, thereby eliminating slower-converging components and speeding up convergence. Empirical results show that quadratic extrapolation can significantly speed up PageRank convergence, though not enough for truly personalized PageRank. The ideas provide a useful approach for accelerating a whole class of iterative methods.