Approximating Aggregate Queries about Web Pages via Random Walks

We present a random walk as an eÆcient and accurate approach to approximating certain aggregate queries about web pages. Our method uses a novel random walk to produce an almost uniformly distributed sample of web pages. The walk traverses a dynamically built regular undirected graph. Queries we have estimated using this method include the coverage of search engines, the proportion of pages belonging to .com and other domains, and the average size of web pages. Strong experimental evidence suggests that our walk produces accurate results quickly using very limited resources.