Vocabulary

Namespace doc: is https://ns.flur.ee/doc#. Structure elements additionally use DoCO (doco:, http://purl.org/spar/doco/), the pattern ontology it extends (po:, http://www.essepuntato.it/2008/12/pattern#), NIF (nif:, http://persistence.uni-leipzig.org/nlp2rdf/ontologies/nif-core#), Dublin Core (dcterms:, http://purl.org/dc/terms/) and FOAF (foaf:, http://xmlns.com/foaf/0.1/) for the document root’s class. The structure terms are those of the parsing engine and are documented in full in its vocabulary reference; the terms below are what the ingest adds.

Classes

Class	Meaning
`doc:Chunk`	A retrieval unit cut from a document’s structure.
`doc:SourceDocument`	The file a graph was extracted from; the `doc:sourceDocument` target.
`doc:Mention`	A span of a chunk naming an entity. Also typed `nif:RFC5147String`.
`doc:Relation`	One relation the language model reported, reified with its evidence and verdict.
`doc:Entity`	An entity extraction minted because no `--entities` source knew it. Known entities keep their own IRI and get no node here.

Chunk properties

Property	Range	Meaning
`doc:chunkIndex`	integer	Position in the document.
`doc:text`	string	The chunk’s text, whitespace collapsed.
`doc:headerPath`	string	Section titles above it, outermost first, joined with `/`. Absent when there are none.
`doc:sourceElement`	IRI, repeated	The elements it was built from, in order.
`doc:sourceDocument`	IRI	The document. Also on every structure node but a table cell, which is reached through its table.
`doc:embedding`	`@vector`	Its embedding, when produced.

Document properties

Property	Range	Meaning
`doc:fileName`	string	The file’s name.
`doc:relativePath`	string	Path relative to the ingested directory, `/`-separated.
`doc:sha256`	string	Hex digest of the bytes.
`doc:mediaType`	string	The media type inferred from the extension.
`doc:byteSize`	integer	Size in bytes.
`doc:pageCount`	integer	Pages, for sources that have them; `0` otherwise.
`doc:escalatedCrops`	integer	Crops a vision model read for this extraction.
`doc:parserRevision`	string	The fluree-doc-parse revision that produced the structure graph.
`doc:chunkCount`	integer	Chunks written.
`doc:chunking`	string	`min/max` characters the chunks were cut with.
`doc:embeddingModel`	string	The embedding model, when one ran.
`doc:embeddingDimensions`	integer	Its vector width.
`doc:ingestedAt`	`xsd:dateTime`	When this extraction was written.
`doc:extractionModel`	string	The language model that extracted, when one ran.
`doc:extractionFingerprint`	string	Hash of the ontology, entity sources, model, guidance and relation mode.
`doc:mentionCount`	integer	Mentions written.
`doc:entityCount`	integer	Distinct entities mentioned, known and new.
`doc:relationCount`	integer	Relations written, every verdict included.

Mention properties

Property	Range	Meaning
`nif:beginIndex`, `nif:endIndex`	integer	Character offsets into the chunk’s `doc:text`.
`nif:anchorOf`	string	The text as written.
`nif:referenceContext`	IRI	The chunk.
`nif:entity`	IRI	The entity, under the IRI its source gave it.
`doc:sourceElement`	IRI	The structure element the span sits in.
`doc:extractedBy`	string	`gazetteer` or `llm`.
`doc:sourceDocument`	IRI	The document.

Relation properties

Property	Range	Meaning
`rdf:subject`, `rdf:predicate`	IRI	The statement.
`rdf:object`	IRI or literal	The statement’s object: an entity, or the literal the model gave when the object was a value or named no entity in the chunk.
`rdfs:label`	string	`subject
`doc:excerpt`	string	The text supporting it, as the model quoted it.
`doc:verdict`	string	`valid`, `repaired` or `rejected`.
`doc:assertionMode`	string	`asserted`, `hedged`, `attributed` or `negated`: how the source states it. Present only when a custom `--system-prompt` asked for it. Not a confidence score and not ordered — see Assertion mode.
`doc:asserted`	boolean	Whether a direct edge was written for it.
`doc:originalPredicate`, `doc:repairNote`	string	On a repaired relation: what the model wrote, and how it was resolved.
`doc:rejectionReason`	string	On a rejected relation.
`doc:sourceChunk`, `doc:sourceDocument`	IRI	Where it was found.

Minted entity properties

Property	Range	Meaning
`schema:name`	string	The canonical name.
`skos:altLabel`	string, repeated	Other surface forms seen.
`doc:nerLabel`	string	Coarse label: PERSON, ORG, GPE, LOC, FAC, EVENT, PRODUCT, WORK_OF_ART, LANGUAGE, MISC, CONCEPT.
`doc:offModel`	boolean	Present and true when the model’s class for the entity is not in the ontology; the node carries no class.
ontology datatype properties	literal	Attributes the model stated and the ontology admits.

Structure terms used by the chunker

The chunker starts at the foaf:Document root and follows po:contains, visiting a node’s po:containsAsHeader child first. It takes text from doco:Title and doco:Paragraph (nif:isString, else rdfs:label), turns each doco:Section’s doco:SectionTitle and doc:sectionLevel into the header path, and reads each doc:TableCell as doc:cellValue with its doc:rowHeader and doc:columnHeader. Every other node is a container it walks through. See What gets built for the rest of the structure graph.

Keyboard shortcuts

Fluree DB