Is authority the same as PageRank?

No. Both use link structure, but HITS explicitly maintains separate hub and authority scores and uses a different computation.

Can one page be both a hub and an authority?

Yes. Its incoming and outgoing links can support both roles.

Does a low score mean a page is unreliable?

No. It may be new, isolated or outside the selected graph, even if its information is excellent.

AI Glossary / Graphs and symbolic AI / Intermediate

Authority

Definition

In link analysis, an authority is a page or node that receives strong endorsement through incoming links from useful hub nodes. In the HITS algorithm, authority is a computed graph score, not a guarantee of factual accuracy, official status or permission to make decisions.

What it means

Some pages primarily provide information; others help people find information by collecting links. HITS captures this relationship with two scores. An authority score reflects links arriving from strong hubs, while a hub score reflects links going to strong authorities.

The two roles can coexist on one page. A useful technical guide may provide substantial information and also link to other resources. The algorithm does not require every page to belong permanently to only one category.

The meaning is specific to the selected graph. A page can be prominent in one topical collection and unimportant in another. This differs from everyday authority, which might refer to expertise or institutional responsibility, and from software authorization, which concerns permitted actions.

How it works

The process begins with a directed graph whose nodes represent pages and whose edges represent links. In query-oriented HITS, the selected graph is assembled around the topic of interest rather than necessarily covering the entire web.

Scores are updated repeatedly. Each page's authority score receives contributions from the hub scores of pages linking to it. Each page's hub score receives contributions from the authority scores of pages it links to. Normalization prevents magnitudes from growing without bound while preserving the relative pattern of scores.

This mutual reinforcement is more informative than simply counting incoming links, because different linking pages can contribute different amounts. However, the result depends on graph construction, link interpretation and numerical details. A link may be a criticism, a navigation shortcut or an automatically generated reference. The algorithm's endorsement interpretation is a modeling assumption, not a direct observation of trust.

Three practical examples

A research portal uses incoming links from curated topic directories as one signal when identifying useful introductory resources.
An internal knowledge search tool distinguishes pages that explain a procedure from index pages that help employees navigate related procedures.
A learning application demonstrates how changing links changes authority rankings, helping students understand the limits of popularity-based signals.

Think of it this way

Think of several reading lists pointing toward books. A book recommended by particularly useful lists may receive more attention than one mentioned by many irrelevant lists.

The analogy stops at the scoring mechanism. A reading list can be mistaken, copied or manipulated, and a frequently recommended book can contain errors. Graph authority helps locate candidates for inspection; it does not replace inspection of their claims.

Worked example

One round of mutual reinforcement

Consider four invented pages. Hub H1 links to resource A and resource B. Hub H2 links only to A. Begin with hub scores of 1 for both H1 and H2.

The first unnormalized authority update gives A a score of 2 and B a score of 1. Using those values, the next hub update gives H1 a score of 3 because it links to both resources, and H2 a score of 2 because it links only to A.

Another authority update would therefore give A 3 + 2 = 5 and B 3. These are illustrative raw updates, not final normalized scores. The example shows why the procedure is more than a permanent count of incoming links. It also shows no evidence that A is factually better: only the link structure has been examined.

Illustrative case study

Fictional case: Maple Manuals applies link analysis to its employee documentation. A retired setup guide receives many links from old department index pages, while its replacement has few links because the indexes were never updated.

The retired guide obtains a strong authority signal. The search team notices that employees still need publication status, version information and ownership checks before relying on a result. It repairs obsolete links and treats the graph score as one ranking input alongside those checks.

The lesson is that graph prominence can preserve yesterday's documentation habits. A mathematically correct calculation may still support an unsuitable recommendation when the graph contains outdated operational relationships.

Common misconceptions and mistakes

Authority means the source is true. The score concerns links, not independent verification of statements.
The page with most incoming links must always win. HITS weights contributions through hub scores.
Hub and authority scores are interchangeable. They summarize different directions of connection.
An authority score grants access or approval power. Security permissions are a separate concept.
A ranking remains valid after the graph changes. New pages, removed links and topical scope can alter the result.

Business use

Authority scores can support discovery in document collections, resource directories and search systems. They are most useful when the link graph contains meaningful relationships and users understand what the score represents.

Before deployment, inspect which kinds of links dominate and whether duplicated templates or obsolete navigation create misleading reinforcement. Evaluate search usefulness with realistic tasks, including cases where the newest correct document has few links. Keep content validity, access restrictions and operational approval separate from graph ranking so that prominence does not become an accidental permission mechanism.

Check your understanding

A newly published official procedure has fewer incoming links than an obsolete guide. Which should a business follow?

Answer: The graph score alone cannot answer that. Staff should establish the applicable approved version. Link analysis can help find documents, but authority in its computational sense is not policy authority.

Related terms

Hub
Directed graph
Information retrieval
Ranking
HITS
link analysis
PageRank
centrality

Learn next

Draw two index pages and three resource pages, then connect them with directed links. Perform one authority update and one hub update by hand. Add a copied index page and observe what changes. Next, study how graph selection and link duplication influence ranking before treating any score as a quality assessment.

Ranking

Sources

Manning, Raghavan and Schütze, Introduction to Information Retrieval, Hubs and Authorities, explains HITS and its mutually reinforcing scores: https://nlp.stanford.edu/IR-book/html/htmledition/hubs-and-authorities-1.html

The companion section explains selecting the topic-related web subset: https://nlp.stanford.edu/IR-book/html/htmledition/choosing-the-subset-of-the-web-1.html

The numerical graph and fictional documentation case are original illustrations.

Previous: Author name disambiguation Next: Authorship attribution