An unambiguous reference to the resource within a given context.
Recommended practice is to identify the resource by means of a string conforming to an identification system. Examples include International Standard Book Number (ISBN), Digital Object Identifier (DOI), and Uniform Resource Name (URN). Persistent identifiers should be provided as HTTP URIs.
Of the fifteen Dublin Core elements, dc:identifier is the one machines care about most. Titles change between editions and names get spelled three ways, but a good identifier lets two systems agree that they are talking about the same object. The definition asks for an unambiguous reference within a given context, and the second half of that phrase is where most of the practical work lies.
What “within a given context” means
An accession number like MS-2021-088 is unambiguous inside one archive and meaningless anywhere else. An ISBN is unambiguous across the book trade. A DOI or an HTTP URI aims to be unambiguous globally. All of them are valid values; the question is which context the record will travel into.
A record that never leaves the local catalog can live with local numbers. Once it is exposed through OAI-PMH or published as linked data, a bare local string becomes a weak link, since a harvester has no way to know whose numbering system it came from. Prefixing it (local:MS-2021-088) or turning it into a URL under a domain the institution controls fixes that.
Stability is the other half of the job. An identifier that is reassigned after a deaccession, or a URL that changes when the repository software is replaced, breaks every citation and every harvested copy of the record. Persistent identifier systems such as DOI, handle and ARK exist mainly to separate the name of a resource from its current location, so that a migration changes the resolver’s target and leaves the identifier alone.
Identification systems and how to write them
The DCMI comment names ISBN, DOI and URN as examples and adds that persistent identifiers should be given as HTTP URIs. In practice the common schemes look like this:
| Scheme | Typical form in a record |
|---|---|
| DOI | https://doi.org/10.5555/example.4471 |
| Handle | https://hdl.handle.net/20.500.12345/678 |
| ISBN | urn:isbn:9780000000002 |
| ISSN | urn:issn:0000-0019 |
| ARK | https://n2t.net/ark:/12345/x9t38 |
The resolver form of a DOI or handle is the version that links and resolves. The prefixed URN forms of ISBN and ISSN make the scheme explicit in a single string, which matters in simple Dublin Core because there is no attribute available to declare the scheme separately.
Identifier of the thing, or of something else?
The value must identify the resource described by the record, and not a related one. That sounds obvious but goes wrong often in digitization projects. If the record describes a scanned copy, the print book’s ISBN identifies a different resource: the original. It is better recorded as dc:source or, in qualified records, through dcterms:relation or one of its sub-properties.
The same issue arises with landing pages and files. A repository landing page URL identifies the page; the PDF behind it has its own URL. Many repositories accept that dc:identifier points to the landing page because it is the more stable of the two, and record the file location elsewhere.
The dcterms side
In DCMI Metadata Terms, dcterms:identifier carries the same definition and a formal range of rdfs:Literal. That literal range is a quiet surprise for linked-data developers: a URI given as an identifier is a string value, while the resource’s own URI is expressed as the subject of the triples. dcterms:bibliographicCitation is also a sub-property of the identifier properties, which is why a full citation string is technically acceptable as an identifying value.
For anyone browsing the vocabulary hub, identifiers are also the key to joining Dublin Core records with format registries: a file described with dc:format and a stable identifier can be matched against characterization results years after ingest, provided nobody reused the number in between.
Example in RDF
<rdf:Description rdf:about="https://example.org/items/4471">
<dc:title>Field notes on coastal erosion</dc:title>
<dc:identifier>https://doi.org/10.5555/example.4471</dc:identifier>
<dc:identifier>urn:isbn:9780000000002</dc:identifier>
<dc:identifier>local:MS-2021-088</dc:identifier>
</rdf:Description>
Questions about dc:identifier
What is dc:identifier used for?
dc:identifier gives an unambiguous reference to the resource being described, such as a DOI, an ISBN, a handle, a URN or a stable web address. It lets systems match, deduplicate and link records that describe the same thing.
Should a DOI be written as a URL in dc:identifier?
The DCMI usage comment says persistent identifiers should be provided as HTTP URIs, so https://doi.org/10.5555/abc is the preferred form. The bare 10.5555/abc string still works for humans but is less reliable for automated linking.
Can a record have several identifiers?
Yes. A digitized book can carry an ISBN for the print edition, a DOI for the digital copy and a local accession number. Each goes in its own dc:identifier statement, ideally in a form that shows which scheme it belongs to.
Is a file checksum a valid identifier?
A checksum such as SHA-256 identifies a specific sequence of bytes, so it can work as an identifier for a file instance. Most repositories keep fixity values in technical or preservation metadata, where their algorithm and date can also be recorded.
How does dc:identifier relate to dcterms:identifier?
Both share the definition and the same usage comment. dcterms:identifier is a sub-property of dc:identifier and declares a literal range, so its values are strings. A URI given as an identifier therefore appears as a literal value there.