A topic of the resource.
Recommended practice is to refer to the subject with a URI. If this is not possible or feasible, a literal value that identifies the subject may be provided. Both should preferably refer to a subject in a controlled vocabulary.
Subject metadata is how most users find resources they did not know existed. Someone researching estuarine sedimentation can find a report titled “Northern estuary survey, 2023” only because a cataloger recorded what the report is about, independent of the words in its title. In DCMI Metadata Terms that statement is dcterms:subject, often written dct:subject, defined in four words: a topic of the resource.
URIs, literals and controlled vocabularies
The usage comment is more prescriptive than the definition. It prefers a URI for the subject, allows a literal when a URI is not feasible, and asks that both refer to a controlled vocabulary where possible. Three levels of practice follow from it.
Concept URIs. The Library of Congress publishes LC Subject Headings as linked data at id.loc.gov, each heading with a stable URI. MeSH, the Getty vocabularies, the FAST faceted headings and many national thesauri do the same. A URI gives the subject a fixed identity: its preferred label can change, translations and alternative labels can be attached, and every record still points to the same concept.
Controlled literals. A heading string from a vocabulary, such as “Sediment transport”, combined with an encoding scheme in qualified Dublin Core. DCMI defines schemes including dcterms:LCSH, dcterms:MESH, dcterms:DDC, dcterms:LCC and dcterms:UDC, so a parser can tell which list a string came from.
Free keywords. Author-supplied tags with no vocabulary behind them. Useful for recall, especially in technical fields where vocabularies lag behind practice, and the hardest to aggregate.
Most collections mix the levels. The example on this page pairs a concept URI from a controlled scheme with an author keyword, which keeps the controlled access point and preserves the author’s own term.
SKOS and local vocabularies
When no published vocabulary fits, a local one can still be built properly. SKOS, the W3C Simple Knowledge Organization System, provides the structure: concepts with URIs, preferred and alternative labels, broader and narrower links. Format registries are a typical case; terms such as “raster image”, “audio container” or “database dump” seldom appear in general subject lists, and a small SKOS scheme makes them citable and translatable. The vocabulary hub lists the SKOS properties used for that purpose.
What does not belong in subject
Several kinds of value end up in subject fields by habit and do more harm than good there:
- The resource type. “Dataset”, “Photograph”, “Report” describe what the resource is, which belongs in dct:type.
- Format terms. “PDF” or “TIFF” are formats, not topics, unless the resource is about the format, as with a specification or a guide.
- Place and period alone. A topic can include a place (“Estuaries—Maine” in LCSH form), but if the record needs to say the resource covers Maine in the 1990s,
dcterms:spatialanddcterms:temporalexpress that more precisely. - Repetition of the title. Splitting the title into words and adding them as subjects adds nothing that full-text search does not already find.
Depth and specificity
Library practice favors the most specific heading that covers the content. A report on sediment in one estuary gets “Estuarine sediments” or a narrower term, not “Geology”. General headings are reached through the vocabulary’s broader links, which a good interface can follow automatically. Assigning both the specific and the general term to every record clutters the broader facet with items that are only marginally about it, which is the problem that hierarchies in vocabularies like LCSH were designed to solve. A description can then explain the scope in sentences, where nuance fits better than in a heading.
Example in RDF
@prefix dct: <http://purl.org/dc/terms/> .
<https://example.org/reports/estuary-sediment-2023.pdf>
dct:title "Sediment survey of the northern estuary"@en ;
dct:subject <https://example.org/concepts/estuarine-sedimentation> ;
dct:subject "estuarine sedimentation"@en ;
dct:type <http://purl.org/dc/dcmitype/Text> .
Questions about dcterms:subject
What is dct:subject?
dct:subject, written dcterms:subject in library documentation, records a topic of the resource. Values range from free keywords to URIs of concepts in controlled vocabularies such as LCSH, MeSH, the Getty thesauri or a project's own SKOS concept scheme.
Should dct:subject values be URIs?
The DCMI usage comment recommends a URI, with a literal allowed when a URI is not feasible, and prefers that both come from a controlled vocabulary. URIs avoid spelling variants and let systems pick up labels in other languages.
How does dcterms:subject relate to dc:subject?
dcterms:subject is a sub-property of dc:subject and shares its basic meaning. The dc element is typically filled with keyword strings, while the dcterms property is the usual choice when subjects are expressed as linked-data concepts.
Can classification numbers go in dct:subject?
Yes. DCMI defines encoding schemes for classifications such as DDC, LCC and UDC, and a class number is a legitimate subject value. Spatial and temporal topics may be better placed in dcterms:spatial and dcterms:temporal.
What is the difference between subject and type?
Subject is what a resource is about; type is what kind of resource it is. A dataset about poetry has subject poetry and type Dataset. Mixing them, for example adding Dataset as a subject keyword, makes both facets less reliable.