What the HITS Algorithm Is
HITS (Hyperlink-Induced Topic Search) is a link-analysis ranking algorithm developed by Jon Kleinberg in the late 1990s. Where PageRank assigns each page a single importance score, HITS assigns two complementary scores to every page: authority (is this page a definitive source?) and hubness (does this page point to many good sources?).
HITS was developed at IBM and influenced early-2000s search engines, most notably Teoma and later Ask.com.
What HITS Ranks
HITS ranks web pages within a query-specific subgraph. This is its sharpest difference from PageRank: HITS is query-dependent — it computes rankings over the pages relevant to a particular search topic, not one global ranking of the web.
Core Inputs Used by HITS
- The link structure of a topic-focused subgraph — typically the top results for a query plus the pages they link to and that link to them
- Two scores per page — authority and hub, mutually defined
Like PageRank, HITS does not analyze page content; ranking emerges from links alone.
How HITS Scores Are Calculated (High-Level)
The algorithm runs a mutual reinforcement loop:
- Every page starts with equal authority and hub scores.
- Authority update — a page’s authority becomes the sum of the hub scores of pages linking to it.
- Hub update — a page’s hub score becomes the sum of the authority scores of pages it links to.
- Repeat until scores converge. The loop is mathematically equivalent to computing dominant eigenvectors of matrices derived from the link graph.
The intuition: good authorities are pointed to by good hubs, and good hubs point to good authorities — a directory page and a definitive source are valuable in different ways.
Conceptual model: PageRank asks “who do important pages endorse?”; HITS asks two questions — “who curates this topic?” and “who do the curators trust?”
Update Frequency
HITS is computed per query, at search time (or precomputed per topic cluster), not as a standing global index — a key practical difference from PageRank.
Known Limitations and Criticisms
- Topic drift — the expanded subgraph can be captured by a densely linked off-topic community, dragging results away from the query
- Computational cost — per-query iteration is expensive at search-engine scale
- Spam vulnerability — mutually linking farms can inflate hub and authority scores
- Query dependence as a cost — no reusable global ranking; every new query recomputes
Where HITS Is Used
HITS-style analysis appears in:
- Historical web search (Teoma/Ask.com)
- Social network analysis — finding brokers (hubs) and leaders (authorities)
- Citation and recommendation network research
- Teaching, as the canonical counterpart to PageRank in link-analysis courses
Summary
HITS introduced the idea that importance has two roles — curators and sources — and that ranking can be tailored per topic rather than computed once for the whole web. PageRank’s global simplicity won the search-engine era, but the hub/authority insight persists wherever networks are analyzed for influence.
References and Sources
- Kleinberg, J. M. Authoritative Sources in a Hyperlinked Environment (1999).
- Wikipedia. HITS algorithm.
- Langville, A. N., & Meyer, C. D. Google’s PageRank and Beyond.