Knowledge Graph
Describe a governed graph dataset through its semantic bindings, qualified assertions, provenance, revisions and permitted use.
Bundle → Layer → Finding → Questions Filled
3 bundles · 3 layers · 5 findings · 11 questions
Schema and scope What the graph is about and which vocabulary it uses.
Ontology binding
The ontologies and vocabularies that give the graph meaning.
Bound vocabularies
The ontologies, schemas and versions the graph uses.
- Which ontologies or vocabularies, in which versions, does this graph use?
- Are there classes or properties in the graph that no bound vocabulary defines?
Graph scope
The domain and coverage the graph claims.
- Which domain, entities and period does this graph claim to cover?
- Who publishes the graph, and under which licence?
Assertions The statements the graph makes.
Statements and identity
Entities, their identifiers and the facts asserted about them.
Entity identity
How entities are identified and reconciled.
- Which identifier scheme does the graph use for entities, and are the identifiers resolvable?
- How are duplicate entities detected and merged?
Assertion provenance
Source, method and time of each assertion.
- Which source and method produced this assertion, and when?
- Is the assertion extracted automatically, inferred or curated by a person?
- Which assertions conflict, and how is the conflict recorded?
Quality and release Whether the graph can be trusted and how it changes.
Validation and versions
Shape validation and release history.
Validation status
Results of shape or constraint checks for a release.
- Which shape or constraint checks were run on this release, and with what result?
- What changed between this release and the previous one?
Classifiers Filled
- Family
- World Models
- Category
- Information and virtual systems
- Entry kind
- entity
- Navigation path
- NAV.INF.DAT.KG
- Domain
- INF.DAT.KG
- Industry
- Cross-industry
- Tags
- knowledgegraphinf.dat.kg
What it is Filled
A knowledge graph is a graph dataset of entities and the typed relationships between them, interpreted through an ontology or schema and carrying provenance for its assertions. It is a dataset with semantics; the ontology it uses and the software that stores it are separate things.
In scope
- Graph identity, coverage and explicit semantic profile
- Local assertion membership, contextual qualification, identity-link decisions and evidence pointers
- Snapshot validation receipts, release references, access conditions and withdrawal records
Out of scope
- Ontology authoring and global concept lifecycle
- Master records and real-world lifecycle of entities described in assertions
- Execution of pipelines, reasoners, query engines, access engines and deletion services
- Universal truth certification, automatic personal-identity merging and unrestricted publication
- Complete property-graph, GQL, RDF 1.2 or embedded-triple mappings
Why it exists Filled
Describe a governed graph dataset through its semantic bindings, qualified assertions, provenance, revisions and permitted use.
Distinguishing features Filled
- A knowledge graph is a dataset; the ontology it uses is a model registered separately.
- Assertions carry provenance, so the graph can say who claimed a fact and how it was derived.
- Differs from a relational dataset by typed links between identified entities that can cross sources.
- Distinct from the graph database software that stores it.
What robots and AI may and may not do Filled
Must not
- Add assertions without provenance or present inferred facts as curated ones.
- Merge entities about people on weak matches without review.
- Assert sensitive facts about identifiable people beyond the graph's lawful purpose.
- Silently change the meaning of a bound ontology term in place.
- Republish the graph against its licence.
Only with a human decision
- Approving changes to the ontology bound to the graph.
- Accepting merges of entities that describe people.
- Publishing a release with personal or sensitive data.
May
- Query the graph within its licence and access conditions.
- Propose new assertions with source, method and confidence recorded.
- Run shape validation and report violations.
- Flag conflicting or unsourced assertions for curation.
Moral aspects Filled
- Linking data about people across sources can reveal more than any source did alone.
- Errors propagate quickly when many systems and AI agents treat the graph as ground truth.
- Bias in sources and extraction shapes what the graph says about groups and places.
Who is affected
- People described in the graph
- Publishers and curators of the graph
- Systems and users that rely on its assertions
Owners Filled
Steward
The data steward or knowledge engineer who answers for the graph's content, scope and releases.
Roles
- Data steward
- Own graph scope, correction decisions and external master bindings.
- Semantic curator
- Review term bindings and identity-link evidence without changing external definitions.
- Evidence reviewer
- Assess source limitations, evaluation completion and disputed claims.
- Release approver
- Authorize local release disposition and consumer-impact handling.
- Access and retention custodian
- Approve purpose-limited views and delegated disposal conditions.
- Consumer
- Respect revision pins, usage conditions and result limitations.
Master systems
- Graph database or triple store
- Data catalogue
- Ontology registry
Links to other meta-models Filled
references
- WM-KNW-001 - Proposed binding to ontology revisions; registry parent is not accepted as semantic inheritance.
- WM-DAT-004 - Shapes and data-contract revision references; external definition lifecycle remains external.
- WM-DAT-005 - Import activity and pipeline run references without local execution ownership.
- WM-DAT-006 - Lineage evidence references with assertion-scoped bindings only.
- WM-DAT-007 - Evaluation reports and scoped outcomes; no evaluation engine implementation.
aligned
- WM-DAT-001 - Proposed same-master graph specialization of a dataset; mapping and version pin remain to be validated.
- RDF 1.1 / SPARQL 1.1 - Baseline graph and query semantics; implementation conformance is untested.
- DCAT 3 / PROV-O / SHACL - Conceptual metadata and evidence alignment, not a completed executable mapping.
neighbor
- WM-KNW-001 Model / Ontology - Registry parent denotes conceptual ancestry only. Reference the applicable ontology revision; do not inherit its definition-authoring lifecycle or infer that the graph is itself an ontology.
- WM-DAT-001 Dataset - Align graph-specific dataset identity with the dataset master; avoid creating a second authoritative identity when both views describe the same dataset.
- WM-DAT-004 Data Schema / Data Contract - Reference shapes and contract revisions; keep graph-specific validation scope and receipts locally.
- WM-DAT-005 Data Pipeline and WM-DAT-006 Data Lineage - Reference import activities and lineage records; assertion-level evidence bindings do not own scheduling, transformation execution or lineage-system lifecycle.
- WM-DAT-007 Data Quality Evaluation - Reference external evaluations; store graph snapshot, metric scope and local release disposition rather than duplicate the evaluation engine.
- Described entity masters and storage/query services - A node reference, graph distribution or endpoint is not the graph owner and does not transfer authority over its referent.
parent
- WM-KNW-001
What else AI and robots need to interact with it Filled
Identity and identifiers required Filled
- A graph is identified by a persistent identifier or dataset URI with a release version.
- Entities inside it are identified by IRIs, often linked to external authority identifiers.
Direct properties not applicable Not applicable
Not applicable
A knowledge graph is information, not a physical object; size, coverage and quality figures are dataset metadata recorded in the structure.
Recognition optional Filled
- Marked by typed links between identified entities and a declared ontology or schema.
- Often confused with the ontology alone, with a graph database product or with a plain linked dataset.
Capabilities and actions required Filled
- Answers queries that follow relationships across entities and sources.
- Supports inference from ontology axioms where a reasoner is used.
- Supports validation of structure against declared shapes.
Hazards and failure modes required Filled
- Wrong merges of entities spread errors across every linked fact.
- Re-identification of people from combined assertions.
- Unsourced assertions mistaken for verified knowledge.
Standards and interfaces required Filled
- W3C RDF 1.1 and SPARQL 1.1 for data and queries.
- W3C OWL 2 and SKOS for ontologies and vocabularies.
- W3C SHACL for shape validation and PROV-O for provenance.
- ISO/IEC 39075 GQL for property graph queries.
Context of use required Filled
- Used for search, data integration, research and grounding of AI systems.
- Personal data in the graph falls under data protection law such as the GDPR.
Sources Filled
- RDF 1.1 Concepts and Abstract Syntax - World Wide Web Consortium
- RDF 1.1 Semantics - World Wide Web Consortium
- Shapes Constraint Language (SHACL) - World Wide Web Consortium
- PROV-O: The PROV Ontology - World Wide Web Consortium
- SPARQL 1.1 Query Language - World Wide Web Consortium
- Data Catalog Vocabulary (DCAT) - Version 3 - World Wide Web Consortium
- RFC 3986: Uniform Resource Identifier (URI): Generic Syntax - Internet Engineering Task Force
- Guide to Attribute Based Access Control (ABAC) Definition and Considerations - National Institute of Standards and Technology
- Knowledge Graphs, Hogan et al., ACM Computing Surveys, 2021
Open questions
- Restore independent external review before canonical or publishable-draft promotion.
- Complete exact source/version and errata checks, including the full access-control publication where stronger policy claims are needed.
- Build and test revision-pinned mappings with blank-node collisions, reversed identity links, failed validation, stale inference, filtered queries and lawful disposal cases.
- Research property-graph, GQL and newer RDF bindings without assuming semantic equivalence to this RDF 1.1 baseline.
- No executable nested schemas, RDF mapping, instance validator or adversarial runtime fixtures.
- No complete property-graph, GQL, RDF 1.2, embedding/vector or probabilistic-graph profile.
- No performance, availability, distributed consistency or reasoner completeness guarantees.
- No jurisdiction-specific privacy/retention/legal analysis or source-license adjudication.
- Independent external review absent; latest revisions, errata and direct HTTP response status remain unverified.
Machine files
Provenance
world-models research · reviewable-draft
Built from: models/wm-dat-011-knowledge-graph/spec.yaml, ver-cy/world-models/card-supplements/wm-dat-011-knowledge-graph.json