skosdex

name	skosdex
description	Search and query skosdex (skosdex.fly.dev) — a skills-first index of ~648 openly-licensed SKOS concept schemes (controlled vocabularies / thesauri) from around the world: AGROVOC, EuroVoc, GEMET, LCSH, Getty AAT/ULAN, IPTC NewsCodes, Loterre, Finto/YSO and many more, ~2.5M concepts with cross-vocabulary skos:exactMatch / closeMatch mappings. Two surfaces: a Solr full-text index (the fast, populated search surface) and a SPARQL 1.1 endpoint (Oxigraph) where each scheme is its OWN NAMED GRAPH — the default graph is empty, so SPARQL patterns must use GRAPH ?g { … }. Third-party, NOT UK Parliament. Use when you need to look up a subject term / concept, walk a broader/narrower thesaurus hierarchy, or bridge a term across controlled vocabularies (e.g. tag parliamentary material against EuroVoc or GEMET).
license	Per-scheme; the distributed bundle is open / public-domain / copyleft only (noncommercial schemes are metadata-only). Each concept carries its scheme's licence — check the manifest.
metadata	{"facility":"skosdex","base-url":"https://skosdex.fly.dev","provenance":{"tier":3,"operator":"skosdex (SKOS aggregator; danbri/skosdex)","service":"skosdex.fly.dev","upstream-data":"~648 third-party SKOS concept schemes, normalised to N-Quads and canonicalised with W3C RDFC-1.0; bundled into a Solr index + an Oxigraph SPARQL store","citation-short":"via skosdex (skosdex.fly.dev)","citation-formal":"skosdex SKOS index (skosdex.fly.dev), retrieved {date}; underlying concept from {scheme}","confidence":"authoritative-passthrough","confidence-notes":"skosdex re-publishes third-party vocabularies without semantic modification (N-Quads syntax normalisation only). Authority for any concept rests with its source scheme, not skosdex — cite the scheme too. Not UK Parliament data."}}

skosdex — SKOS thesaurus index

Hosted demo: https://skosdex.fly.dev/ · Source: https://github.com/danbri/skosdex

This is a third-party facility, not a UK Parliament one. skosdex collects, normalises (N-Quads + W3C RDFC-1.0 canonicalisation) and indexes many SKOS concept schemes into one searchable place. As of the 2026-06-12 build: 648 schemes in the bundle, ~2.5M concept documents in Solr, ~3.12M skos:Concept across the SPARQL named graphs. It carries the cross-vocabulary mapping web (skos:exactMatch / skos:closeMatch), so a term in one thesaurus can be bridged to EuroVoc, GEMET, LCSH, Wikidata, etc.

Two query surfaces

Surface	URL	Use it for
Solr	/solr/skos/select	Fast full-text search over labels and definitions. The populated, default surface.
SPARQL 1.1	/query (Oxigraph)	Graph queries: hierarchy walks, mappings, per-scheme slices.

There is also a corpus manifest at /manifest.json (every scheme, its slug, namespace URI, licence, and triple count).

⚠️ The one thing to know about the SPARQL endpoint

The bundle is loaded as one named graph per scheme — the graph IRI is the scheme IRI. The default graph is empty. So a plain

SELECT * WHERE { ?s ?p ?o } LIMIT 5      -- returns NOTHING

You must wrap patterns in GRAPH ?g { … }:

SELECT (COUNT(*) AS ?n) WHERE { GRAPH ?g { ?s a <http://www.w3.org/2004/02/skos/core#Concept> } }
-- => 3121417

This single gotcha accounts for almost every "the endpoint is empty" false alarm. (Contrast with the UK Parliament sparql skill, whose DDP store keeps everything in the default graph.)

Namespaces

PREFIX skos: <http://www.w3.org/2004/02/skos/core#>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
PREFIX dcterms: <http://purl.org/dc/terms/>

Solr fields

Each Solr doc is one concept. Multi-valued unless noted.

Field	Meaning
id	the concept URI (also _root_)
scheme	the concept-scheme URI it belongs to
prefLabel	preferred labels (all languages, mixed)
prefLabel_en, prefLabel_fr, …	per-language preferred labels
altLabel / altLabel_en / …	alternative labels / synonyms (mixed + per-language)
lang	the languages a concept carries (multi-valued)
exactLabel	the lowercased exact-match label index
definition	definition(s) (often a URI to a definition node)
broader / narrower	hierarchy parents / children (URIs)
exactMatch / closeMatch / mapping	cross-vocabulary links (URIs)

Field-scoped Solr queries work: prefLabel_en:climate, altLabel:GDP. Filter to one scheme with fq=scheme:"<scheme-uri>".

⚠️ Bare terms need a field. Solr's configured default field here is not the label text, so a raw q=ocean matches nothing. The parl skosdex search CLI auto-scopes for you — and because the index now has per-language label fields, a bare parl skosdex search "ocean" defaults to English-only, rewritten to prefLabel_en:(ocean) OR altLabel_en:(ocean). Use --lang any to search the language-mixed prefLabel/altLabel, --lang fr for another language, --field a,b for explicit fields, or --raw-q to pass q verbatim.

🌐 Per-language label fields now exist. The index carries prefLabel_en, altLabel_en, prefLabel_fr, … (plus a lang field), so you can search and read back English-only on the Solr surface — q=prefLabel_en:(…), fl=prefLabel_en. The bare language-mixed prefLabel/altLabel fields still exist for cross-language search via --lang any.

Worked examples (curl)

# English-only full-text search (per-language field), two hits
curl -sLG 'https://skosdex.fly.dev/solr/skos/select' \
  --data-urlencode 'q=prefLabel_en:(climate change)' \
  --data-urlencode 'rows=2' \
  --data-urlencode 'fl=id,scheme,prefLabel_en'

# Same via the CLI (bare term auto-scoped, English by default),
# restricted to one scheme:
parl skosdex search "climate change" \
  --scheme http://eurovoc.europa.eu/100141 --rows 2 --fl id,prefLabel_en

# Concept → concept: nearest neighbours via embedding KNN (no model)
parl skosdex similar http://www.eionet.europa.eu/gemet/concept/1471 --k 5 --cross 1

# SPARQL: every label containing "parliament", across all graphs
curl -sL 'https://skosdex.fly.dev/query' -X POST \
  -H 'Accept: application/sparql-results+json' \
  -H 'Content-Type: application/x-www-form-urlencoded' \
  --data-urlencode 'query=
PREFIX skos: <http://www.w3.org/2004/02/skos/core#>
SELECT ?c ?l WHERE { GRAPH ?g {
  ?c skos:prefLabel ?l . FILTER(CONTAINS(LCASE(STR(?l)),"parliament")) } } LIMIT 5'

Semantic matching via embeddings

skosdex publishes concept embeddings (all-MiniLM-L6-v2, 384-dim, L2-normalised) so you can match by meaning, not just label text:

• Concept → concept (no model needed): parl skosdex similar <iri> hits the hosted KNN endpoint /api/similar?id=…&k=…&cross=1 and returns the nearest concepts, optionally across schemes.
• Any text / web page → concepts (needs the model): download the combined space (/embeddings/_all.emb.json metadata + /embeddings/_all.emb.f16 float16 vectors), embed your text with the same model, and take the dot product (= cosine, both sides are unit vectors). Runnable Python recipe in reference.md.

The embedding space is a curated subset (currently ESCO ~19k, GEMET ~5.6k, and the UK Parliament thesaurus ~1.7k — all of it), not the full 2.5M-concept corpus — which makes it well suited to semantically tagging UK Parliament material.

UK Parliament relevance

skosdex is general-purpose, but two bridges matter here:

• Subject tagging against an international vocabulary. EuroVoc and GEMET are both in the corpus; you can map a free-text parliamentary topic to a controlled-vocabulary URI for interoperable tagging.
• The Parliament Thesaurus. The UK Parliament SKOS thesaurus is not on api.parliament.uk/sparql (its Concept class is unpopulated) and natively lives on the legacy linked-data-api (lda.../terms) — but skosdex does carry a partial copy as the named graph http://data.parliament.uk/terms/ (~1,724 concepts, English prefLabels). So you can tag Parliament material against Parliament's own controlled vocabulary here, and cross-walk to EuroVoc / GEMET / LCSH in the same query. Note it is partial (the full thesaurus is far larger) and skews to named entities (organisations, schemes), so phrase-match recall on free text is modest — pair it with EuroVoc/GEMET for topical coverage.

Using the CLI

See ../parl/SKILL.md for the CLI-wide conventions (output modes, flag rules, idiomatic chains).

This skill ships with the same Node CLI as the rest of the repo. From the repo root:

node bin/parl.mjs skosdex --help

Or after npm link:

parl skosdex --help

Examples

parl skosdex search "climate" --rows 5 --fl id,scheme,prefLabel_en

Full-text Solr search (bare term auto-scoped, English-only by default).

parl skosdex search "renewable energy" --scheme http://eurovoc.europa.eu/100141

Same search, restricted to one scheme via a Solr fq filter. Add --lang any to search all languages, --lang fr for French.

parl skosdex similar http://www.eionet.europa.eu/gemet/concept/1471 --k 5 --cross 1

Nearest concepts to a concept IRI via embedding KNN (no model needed).

parl skosdex query 'SELECT (COUNT(*) AS ?n) WHERE { GRAPH ?g { ?s a <http://www.w3.org/2004/02/skos/core#Concept> } }'

SPARQL — remember GRAPH ?g { … }; the default graph is empty.

parl skosdex schemes --limit 20

List the concept-scheme named graphs (graph IRI == scheme IRI).

parl skosdex concept http://aims.fao.org/aos/agrovoc/c_27742

Labels + broader/narrower + cross-vocabulary mappings for one concept.

parl skosdex manifest

The corpus manifest: every scheme with slug, namespace, licence, triple count.

Library use (Node + browser)

import * as skosdex from '../../lib/facilities/skosdex.mjs';
const hits = await skosdex.search('prefLabel:climate', { rows: 5 });

Uses only fetch / URL, so the same source runs in Node 18+ and modern browsers.

Notes & caveats

• Not UK Parliament. Tier-3 third-party. Authority for any concept rests with its source scheme; cite the scheme, not just skosdex.
• Licensing is per-scheme. The distributed bundle is open / public-domain / copyleft only; noncommercial schemes are held as metadata in the manifest's excluded list and are not in the RDF bundle. Check manifest before redistributing anything.
• Solr is the populated surface; SPARQL is the graph surface. If a SPARQL query comes back empty, you almost certainly forgot GRAPH ?g { … }.
• Endpoint is read-only and hosted on a small Fly.io instance — be polite, keep result sets bounded, prefer Solr for label lookups and SPARQL only for graph-shaped questions.

See reference.md for the full Solr field reference, named-graph mechanics, hierarchy-walk and cross-mapping recipes, and the manifest structure.

Provenance to cite

Tier 3 — third-party aggregator (skosdex), authoritative pass-through.

• Inline cite: "(via skosdex)" — once per paragraph — plus the underlying scheme (e.g. "EuroVoc, via skosdex").
• See ../../docs/provenance.md for the cross-skill rules.