Simple Knowledge Organization System (SKOS)

SKOS represents thesauri, taxonomies, subject headings, and other knowledge organization systems on the web. Its concept, concept scheme, label, and broader or narrower relations let publishers share controlled terms without recasting every vocabulary as a formal class ontology. Mapping relations can express links between schemes while preserving their distinct origins and intended scopes.

A broader relation is not automatically a logical subclass statement, and a close match is weaker than exact identity. SKOS does not decide whether a source assigned the right concept to a record. Preserve the vocabulary version, concept URI, original source label, and mapping provenance. This lets a reviewer distinguish a publisher's official hierarchy from a local crosswalk added later.

A fictional agency publishes a drug-class vocabulary in SKOS, and a fictional company maintains its own therapeutic-area list. The company wants to map its terms to the agency's.

For each local term the mapping records a property and a reason. "Oncology" maps to the agency's concept for cancer treatments with an exact match, citing the definitions in both. "Rare diseases" maps to the agency's concept for orphan conditions with a close match, because the definitions overlap without coinciding. "Immunology" is a broader match to two agency concepts. The mapping set is published as its own dataset with an author, a date and a version, separate from both vocabularies. When the agency revises its scheme, each mapping is rechecked against the new version, and the mappings that no longer hold are retired with a date.

A broader relation is not a subclass relation. In a thesaurus, a narrower concept can be a part, an instance or an associated topic, and a reasoner that treats the link as subsumption will infer false statements. If a formal class hierarchy is needed, model it in an ontology language and map between the two.

SKOS is also not a statement about the world. A concept is an idea in a scheme, not a thing, and a document tagged with a concept has been classified by someone, by some process, at some time. The tag is a claim with a source. Classification quality, inter-annotator agreement and drift are separate questions, and they need evidence that the scheme itself does not supply.

Controlled vocabularies are living documents. Concepts are deprecated, merged, split and relabelled, and a good publisher marks these changes in the data, with replacement links and dates. Users who store concept IRIs without the scheme version cannot tell whether a tag from three years ago meant what the concept means now. When a vocabulary is revised, the practical task is to find every record tagged with a changed concept and decide, record by record or by rule, whether the tag stands. That work needs the tags to be stored with their scheme version and the date of assignment, and the decisions to be stored as new assertions, with the old ones kept.