AI Glossary / Data engineering and architecture / Advanced

Linked data

Definition

Linked data is an approach to publishing and connecting structured data using Web identifiers and explicit relationships. It identifies things with URIs, makes useful descriptions available through Web lookup and links those descriptions to other identified things. RDF is central to the conventional approach. Linked data is not simply a collection of web pages with hyperlinks, and it does not automatically mean that the data is openly licensed.

What it means

Two datasets may discuss the same place, organization or concept while using different local names. Shared or explicitly connected identifiers help make those relationships usable across datasets. A program can follow a relationship to learn more about the identified entity instead of relying only on a matching text label. For AI retrieval, this can supply useful context and connections, but the links are still assertions that need appropriate interpretation. A well-connected description can contain mistakes, and two similar names do not prove that their underlying entities are identical.

How it works

Tim Berners-Lee's Linked Data principles call for URIs identifying things, HTTP-based lookup, useful structured information and links to other URIs. Relationships describe how identified resources connect, enabling people and software to explore further information. The meaning of each relationship matters: part of, located in and the same as express different claims. Publishing a link does not independently verify that claim. Maintain identifiers, descriptions and relationship meanings so consumers can understand what the data asserts and where that assertion came from.

Three practical examples

• A museum collection links an artwork's creator identifier to a maintained description of that creator.
• A research dataset identifies a place using a shared identifier rather than depending solely on a place-name string.
• An AI information service follows explicitly described relationships between concepts and sources while retaining evidence about the source of each assertion.

Think of it this way

Imagine a reference book whose entries contain precise cross-references to other entries, including those in other books. A reference explains the connection instead of merely saying that another page exists. Readers can move from a work to its creator and then to related works. Linked data makes comparable connections available in structured form to software. The analogy does not mean that every linked entry agrees with every other one or that a cross-reference is a certificate that the information is correct.

Worked example

Consider an original fictional collection with two artworks and one creator. Artwork A has a created-by link to creator identifier C, and artwork B has the same kind of link to C. There are two artwork entities, one creator entity and two creator relationships. A program can identify the shared creator through C even if one display labels the creator Alex Rivera and another uses A. Rivera. This works because the example stipulates the same identifier for the same entity. Matching those two labels alone would not establish identity, since abbreviated or repeated names can refer to different people. The relationship count is also not the number of distinct creators.

Illustrative case study

A fictional tourism-data team combines descriptions of places from two sources. Both contain a location named Riverside, and an initial matching rule treats the names as identical entities. Reviewers discover that the sources describe different towns. The team corrects the identity mapping and records the identifiers and supporting context used for future links. The scenario illustrates why linked data depends on accurate entity relationships rather than merely adding connections. It does not claim that identifiers eliminate every matching problem or that an AI-generated relationship should be accepted without checking the evidence behind it.

Common misconceptions and mistakes

Linked data is not synonymous with linked open data. Openness concerns reuse conditions as well as the technical form. A URI identifying something also should not be confused with proof that the thing's description is authoritative. Another mistake is using identity links when the intended relationship is merely similarity or association. A graph of connections can spread that mistake into later queries. Finally, publishing structured data does not guarantee that identifiers remain resolvable or descriptions remain current. The approach needs maintenance as well as initial publication.

Business use

The concept helps organizations connect descriptions across catalogs, reference datasets and knowledge services. Decide what each identifier denotes and document the relationships consumers are expected to use. Verify mappings where names are ambiguous, and preserve the source and context of assertions. For AI applications, use those connections as structured evidence rather than as automatic truth. Evaluate whether following a link produces the relevant, current information for the user's question. Technical interoperability is valuable when the linked meanings are also suitable and when consumers can recognize gaps or conflicting descriptions.

People also ask

Is all linked data open data?

No. Linked data can be used within restricted settings, while openness is a separate question.

Do matching names prove two records identify the same thing?

No. Names can be ambiguous or shared.

Does a link establish that its relationship is correct?

No. It records an assertion whose evidence and meaning still matter.

Check your understanding

How many creator entities were in the toy collection?

Answer: One.

How many created-by relationships were present?

Answer: Two.

Why could the program connect the artworks despite different displayed names?

Answer: Both referred to the stipulated same creator identifier.

Why did the fictional Riverside mapping fail?

Answer: A shared name was mistaken for evidence of shared identity.

Related terms

• RDF
• URI
• Semantic Web
• entity resolution
• knowledge graph
• linked open data

Learn next

Sketch a fictional dataset with two works, two creators and one organization. Assign distinct identifiers and label each relationship explicitly. Then introduce a second source using an abbreviated creator name. Explain what additional evidence would justify an identity mapping and how you would preserve uncertainty if that evidence were missing.

More in Data engineering and architecture

• Linda-like systems - Linda-like systems coordinate concurrent processes through a shared tuple space inspired by the Linda model.
• Linear belief function
• Linguistic value - A linguistic value is a word or phrase used as a value of a linguistic variable, such as low, moderate or high for workload.
• Local-first software - Local-first software gives the user's local copy of data a primary role, allowing useful work without depending on a remote server for each operation.
• Locale (computer hardware) - A locale is an abstraction representing a group of computing and memory resources within a parallel system.
• Locks with ordered sharing - Locks with ordered sharing are database concurrency-control mechanisms that permit selected conflicting operations to proceed in a required order while...

Sources

Tim Berners-Lee, Linked Data, Design Issues: https://www.w3.org/DesignIssues/LinkedData.html The artwork calculation, reference-book analogy and Riverside case are original educational illustrations. The design note supports the linking principles and distinction from openness; the entry does not treat its historical implementation examples as current product recommendations or automatic guarantees of data quality.

Previous: Linkage disequilibrium score regression Next: Linking Bias