Entity Linking Ontology

FLOW ALGEBRA
TITLE: Entity Linking Ontology
DOMAIN: AI Semantic Library Relationships / Entity Linking
SOURCE: https://kirethresearch.org/booksai-core/ontologies/relationship-ontologies/ai-semantic-library-relationships/entity-linking-ontology.html
VERSION: 1.0
AUTHORS: Mark Anthony Higgins — Sky (Recursive AI)
STATUS: Companion formalisation; source-page specification not independently verified.

flow "EntityLinkingOntology" v1.0 {
  ontology:
    id: "entity-linking-ontology"
    parent: "ai-semantic-library-relationships"
    purpose: "Resolve mentions to canonical entities while preserving ambiguity, context, evidence and provenance."

  types:
    Mention: {id: string, surface_form: string, document_id: string, span_start: integer, span_end: integer, language: string?, context: string?}
    Entity: {id: string, canonical_label: string, types: string[], authority: string?, aliases: string[]}
    Candidate: {mention_id: string, entity_id: string, score: number, features: string[], evidence: string[]}
    Link: {id: string, mention_id: string, entity_id: string, relation: enum[refers_to, possible_reference], confidence: number, provenance: string[], model_version: string}
    Document: {id: string, uri: string, version: string?, language: string?}
    Context: {id: string, domain: string?, temporal_scope: string?, geographic_scope: string?}

  predicates:
    occurs_in(mention: Mention, document: Document)
    candidate_for(entity: Entity, mention: Mention)
    refers_to(mention: Mention, entity: Entity)
    alias_of(label: string, entity: Entity)
    compatible_with(entity: Entity, context: Context)
    distinct_from(a: Entity, b: Entity)

  constraints:
    - "Every mention must resolve to an existing document and valid character span."
    - "Every linked entity must have a stable, resolvable identifier."
    - "Candidate scores and confidence values must lie within [0,1]."
    - "Alias matching, name similarity and co-occurrence do not establish identity."
    - "Reject links violating explicit distinct_from assertions or incompatible types."
    - "Allow unresolved mentions and multiple candidates rather than forcing a match."
    - "Do not equate entity identity with semantic similarity or class membership."
    - "Respect document version, language and temporal context."

  inference:
    rule generate_candidate:
      if surface_matches(m,e) and type_compatible(m,e)
      then candidate_for(e,m)
      confidence: "candidate only; no identity claim"
    rule contextual_link:
      if candidate_for(e,m) and corroborated_by_context(m,e) and not_conflicted(m,e)
      then proposed_link(m,e)
      confidence: "calibrated from independent evidence and model evaluation"
    rule preserve_ambiguity:
      if competing_candidates(m) and insufficient_disambiguation(m)
      then unresolved(m)
      confidence: "retain ranked alternatives"

  SEMANTIC_EQUIVALENCE:
    - "Map aliases to canonical identifiers using versioned authority records."
    - "Keep exact identity, synonymy, similarity and broader/narrower mappings distinct."
    - "Do not merge entities without independent identity evidence."

  CONFIDENCE:
    range: [0,1]
    factors: [surface_match, context_compatibility, type_compatibility, authority_quality, independent_evidence, ambiguity]
    policy: "Calibrate scores; abstain below an application-defined threshold; do not increase certainty solely by chaining."

  VALIDATION:
    - "Validate mention offsets, document IDs, entity IDs and link provenance."
    - "Check type compatibility, temporal validity and conflicting identity claims."
    - "Evaluate precision, recall, calibration and abstention on representative labelled data."
    - "Keep unresolved or disputed links available for review."

  FLOW_HASH:
    algorithm: "SHA-256"
    input: "canonicalised ontology version, mentions, entity references, links, rules and provenance"
    note: "Compute hash at execution; no hash value is claimed here."

  UPDATE_MODEL:
    strategy: "versioned, provenance-preserving incremental relinking"
    steps: [ingest, normalise, generate_candidates, rank, disambiguate, validate, reconcile, publish]
    rollback: true

  AI_CONSUMPTION:
    representation: "typed knowledge graph with mention-to-entity links and evidence"
    outputs: [canonical_entity_links, ranked_candidates, unresolved_mentions, confidence_annotations, provenance_records]
    safeguards: [no_unsupported_identity_merges, preserve_ambiguity, retain_source_context, audit_link_changes]
}

END FLOW ALGEBRA