EN
Back to the archive

The encyclopedia · R&D & Science · Technical decision · 1997–1999

HITS ranked pages by who links to whom, not by their own words

Kleinberg's HITS scored every page twice — as authority and hub — so mutually endorsing pages rose without describing themselves.

Cornell University

The solution

Searching the early web by text failed on broad queries: Harvard's homepage does not use the word 'Harvard' most prominently, and search-engine homepages do not mention 'search engines'. Jon Kleinberg argued that hyperlinks encode human judgment about authority.

His HITS algorithm divides pages into hubs — pages that point to many good sources — and authorities — pages many hubs point to. Starting from a focused subgraph, it updates both scores in alternation: authorities inherit hub weight, hubs inherit authority weight.

Because the iteration converges to principal eigenvectors of the link matrix, the algorithm is mathematically stable and surfaces small communities of mutually reinforcing pages. Kleinberg first presented it at SODA 1998 and published the journal version in 1999.

Why it worked

  • Links encode human endorsements that text cannot
  • Hub/authority duality captures two different roles
  • Eigenvector iteration converges to stable rankings
  • Broad queries need structure, not just word counts
What it achievedHubs vote for authorities, then get voted backclever

What can be applied

When content says nothing, judge by endorsements — and score the endorsers too. Mutual reinforcement turns many noisy links into a stable ranking, as long as you iterate to convergence.

Aftermath

HITS became a standard link-analysis method alongside PageRank and influenced search, recommendation and citation analysis; the hub/authority vocabulary is now part of web science.

Sources

spotted an error? The archive wants to know.

Related cases