Discovering and Leveraging the Most Valuable Links for Ranking

摘要

On the Web, visits of a page are often introduced by one or more valuablelinking sources. Indeed, good back links are valuable resources for Web pagesand sites. We propose to discovering and leveraging the best backlinks of pagesfor ranking. Similar to PageRank, MaxRank scores are updated {recursively}. Inparticular, with probability $\lambda$, the MaxRank of a document is updatedfrom the backlink source with the maximum score; with probability $1-\lambda$,the MaxRank of a document is updated from a random backlink source. MaxRank hasan interesting relation to PageRank. When $\lambda=0$, MaxRank reduces toPageRank; when $\lambda=1$, MaxRank only looks at the best backlink it thinks.Empirical results on Wikipedia shows that the global authorities are veryinfluential; Overall large $\lambda$s (but smaller than 1) perform best: theconvergence is dramatically faster than PageRank, but the performance is stillcomparable. We study the influence of these sources and propose a few measuressuch as the times of being the best backlink for others, and related propertiesof the proposed algorithm. The introduction of best backlink sources providesnew insights for link analysis. Besides ranking, our method can be used todiscover the most valuable linking sources for a page or Website, which isuseful for both search engines and site owners.