Core Concepts
Revisiting the complexity of ApproxContributions and BiPPR algorithms for efficient PageRank estimation.
Abstract
The content discusses the ApproxContributions algorithm for computing PageRank contributions and its worst-case complexity bounds. It also introduces the BiPPR algorithm, a combination of ApproxContributions and Monte Carlo simulation, for single-node PageRank estimation. The analysis provides insights into the computational complexities and variance of estimators in these algorithms.
ApproxContributions Algorithm:
Introduced by Andersen et al. for computing PageRank contributions.
Worst-case complexity bound: O(nπ(t)/ϵ · min(∆in, ∆out, √m)).
Applications in estimating random-walk probabilities.
BiPPR Algorithm:
Combines ApproxContributions with Monte Carlo simulation.
Computes a multiplicative (1±c)-approximation of π(t) w.p. at least (1−pf).
Expected computational complexity: O(nπ(t)/ϵ · min(∆in, ∆out, √m) + nr).
Variance Analysis:
Variance of estimator ˆπ(t) in BiPPR is bounded by ϵnr · π(t).
Median Trick Optimization:
Boosts success probability to (1-pf) using reruns and median trick.
Simplicity vs Complexity:
Comparison with previous complex analyses highlights simplicity in deriving tight upper bounds.
Stats
We give a worst-case complexity bound of ApproxContributions as O(nπ(t)/ϵ · min(∆in, ∆out, √m)).
The expected computational complexity of running BiPPR is O(nπ(t)/ϵ · min(∆in, ∆out, √m) + nr).
Quotes
"ApproxContributions has become a cornerstone for computing random-walk probabilities."
"Our results show that the simple ApproxContributions algorithm is already optimal."