Can a page be both a hub and an authority?

Yes. The framework assigns separate scores, and a page may link to useful sources while also receiving links from useful hubs.

Does every node with many outgoing links have a high HITS hub score?

Not necessarily. The authority scores of those destinations and the selected graph matter.

Is a hub score a credibility rating?

No. It is a link-structure measure that may inform retrieval but does not independently establish the truth of content.

AI Glossary / Graphs and symbolic AI / Intermediate

Hub

Definition

A hub is a node that connects to other nodes in a way that makes it useful for navigating a network. In the HITS link-analysis framework, a good hub points to high-scoring authorities, while good authorities receive links from high-scoring hubs. The term's precise meaning depends on the network model and scoring method.

What it means

In everyday network descriptions, hub may simply mean a highly connected node. In web-search theory, the HITS meaning is more specific: a page can be a useful guide because it links to valuable sources, even if it is not itself the main source of the information.
This entry focuses on that hub-and-authority interpretation, consistent with the term's AI and graph context. It is different from a networking hardware hub or a branded software hub. A node can receive both a hub score and an authority score; the two roles are not mutually exclusive categories.

How it works

Represent pages and directed hyperlinks as a graph. Maintain a hub score and an authority score for each page. In the core HITS updates, a page's authority score is based on the hub scores of pages linking to it, while its hub score is based on the authority scores of pages it links to.
Repeat the updates with normalization so the scores do not merely grow in scale. The result depends on the selected graph and the procedure's assumptions. A topic-focused collection of pages can yield different scores from a broader graph. These are structural scores, not independent verification of the linked content's truth or quality.

Three practical examples

A teaching web graph distinguishes a curated resource list from the technical articles to which it links.
An information-retrieval experiment compares raw outgoing-link counts with hub scores that account for the linked pages' authority scores.
A documentation analysis explores which index pages connect users to strongly interconnected reference material, while checking usefulness separately with actual tasks.

Think of it this way

Think of a knowledgeable reading-list compiler who points readers toward valuable sources. Their value lies partly in the quality of the collection, while the sources provide the substantive material. The analogy is limited because HITS infers scores from links, not from directly assessing expertise, editorial care, truthfulness, or whether each link is a recommendation.

Worked example

Construct a toy directed graph with four pages: H1, H2, A, and B. H1 links to A and B. H2 links only to A. A and B have no outgoing links in this toy graph. Start by assigning hub score 1 to every page.
One authority update gives A a value of 2 because H1 and H2 point to it, and B a value of 1 because only H1 points to it. H1 and H2 receive authority value 0 because no page points to them. Using these unnormalized authority values, the next hub update gives H1 a value of 2 + 1 = 3 and H2 a value of 2. A and B receive hub value 0.
If the hub vector is normalized to Euclidean length 1, the nonzero scores become 3 divided by the square root of 13, approximately 0.832, and 2 divided by the square root of 13, approximately 0.555. This is one illustrative update cycle, not the final converged ranking. Normalizing the intermediate authority vector would scale both hub values equally and leave their ratio unchanged.

Illustrative case study

In a fictional documentation project, Harbor Guides finds that an index page has many outgoing links but includes several unrelated sections. A reviewer compares its link count with a topic-specific hub analysis and then manually checks whether the linked material helps readers complete the intended task.
The team treats the score as one structural signal and records the graph-selection procedure. It does not promote the page as factually authoritative solely because it has a high hub score. The invented case illustrates interpretation and review rather than a measured improvement in search quality.

Common misconceptions and mistakes

A hub score is not just an outgoing-link count, although counts affect early updates in a simple initialization. The scores of linked authorities matter. A page with many irrelevant or weakly supported links is not automatically the best hub for a topic.
Do not confuse incoming authority with outgoing hub value. Nor should a high score be treated as proof of factual reliability, neutrality, or human expertise. Link patterns can reflect navigation, duplication, or manipulation rather than endorsement. Scores also depend on which pages and edges are included, so a ranking is conditional on the constructed graph.

Business use

The concept helps explain link-based information retrieval and the different roles of sources and collections of sources. It can be useful when exploring documentation structures, reference networks, or curated directories where links carry meaningful relationships.
Define the topic, graph boundary, and meaning of an edge before interpreting a hub ranking. Compare structural scores with content relevance and user needs. Keep hub and authority labels distinct in reports, and explain whether a reported value is a raw count, an intermediate update, or a normalized result from a completed algorithm. This prevents a technical score from being mistaken for a broad endorsement.

Check your understanding

Why does H1 receive unnormalized hub value 3 after the worked authority update?

Answer: It links to A with authority value 2 and B with authority value 1, so its hub update sums those values to 3.

Related terms

Authority
Directed graph
Information retrieval
HITS
hyperlink analysis
centrality
PageRank
outgoing link

Learn next

Repeat the toy updates and observe how normalization changes scale without changing a cycle's ratios. Then add a link from A to another page and trace its effect. Compare HITS hub scores with degree counts and explain the different questions they answer.

Sources

Sources: Jon Kleinberg, Structural Analysis of the World Wide Web: https://www.w3.org/1998/11/05/WC-workshop/Papers/kleinber1.html Manning, Raghavan, and Schütze, Introduction to Information Retrieval, Hubs and Authorities: https://nlp.stanford.edu/IR-book/html/htmledition/hubs-and-authorities-1.html

Previous: HTTPRange-14 Next: Huber regression

Hub