Controlled Vocabulary

A predefined, finite set of terms that represent the only permissible values for a given metadata field or indexing operation. This eliminates the noise of synonyms, jargon, and spelling variants.

• Enforcement: Systems reject any term not present on the list
• Example: A product database allowing only 'Smartphone', 'Tablet', or 'Laptop'—not 'Cell Phone' or 'Mobile Device'
• Purpose: Guarantees that every piece of content is tagged with the exact same label for the exact same concept

A set of equivalent terms where one is designated as the preferred term and the rest are non-preferred variants. When a user searches for any synonym, the system maps it to the preferred term for retrieval.

• Mapping: 'Automobile' and 'Car' both resolve to the preferred term 'Motor Vehicle'
• Invisible to user: The mapping happens at the index level, not the interface level
• Critical for recall: Ensures a search for 'heart attack' also retrieves documents tagged with 'myocardial infarction'

A parent-child taxonomy that organizes terms into broader and narrower categories, enabling faceted navigation and query expansion.

• Broader Term (BT): 'Canine' is broader than 'Golden Retriever'
• Narrower Term (NT): 'Golden Retriever' is narrower than 'Canine'
• Polyhierarchy: A term like 'Electric Vehicle' can legitimately sit under both 'Vehicles' and 'Sustainable Technology'
• Use case: A search for 'fruit' automatically expands to include documents tagged with 'apple', 'banana', and 'citrus'

A non-hierarchical 'See Also' link between two terms that are conceptually related but not in a parent-child lineage. This captures semantic connections that a strict taxonomy misses.

• Related Term (RT): 'Welding' is related to 'Metallurgy'
• Related Term (RT): 'Inflation' is related to 'Interest Rates'
• Purpose: Guides indexers and searchers to adjacent concepts they may not have considered
• Implementation: Often encoded in thesauri conforming to the ISO 25964 standard for interoperability

A precise textual definition that explicitly delimits the meaning of a term within the specific context of the vocabulary, preventing misinterpretation by human indexers.

• Disambiguation: 'Mercury' (planet) vs. 'Mercury' (element) vs. 'Mercury' (mythological figure)
• Usage guidance: 'Use this term only for accidental fluid spills. For intentional releases, use Discharge.'
• Historical context: 'This term was deprecated in 2022 and replaced by Machine Learning Operations.'
• Essential for governance: Scope notes are the contract between the vocabulary designer and the content tagger

Every term in a controlled vocabulary is assigned a machine-actionable, immutable URI that remains constant even if the human-readable label changes over time.

• Decouples label from concept: The label 'Swaziland' can update to 'Eswatini' without breaking all existing metadata references
• Format: Often a UUID or a resolvable HTTP URI like https://schema.org/Product
• Linked Data foundation: These URIs enable the vocabulary to function as a node in the broader semantic web, allowing external systems to unambiguously reference the exact concept

A hierarchical classification scheme that organizes concepts into parent-child relationships, providing a controlled vocabulary that structures a domain's information architecture. Unlike flat term lists, taxonomies enforce inheritance and subsumption logic.

• Defines broader/narrower term relationships
• Enforces consistent content categorization
• Foundation for faceted navigation and filtering

A formal, explicit specification of a shared conceptualization that defines the types, properties, and interrelationships of entities within a domain. Ontologies go beyond hierarchical taxonomies by enabling logical inference and semantic reasoning through property constraints and class axioms.

• Supports transitive, symmetric, and inverse relationships
• Enables automated consistency checking
• Expressed in languages like OWL and RDFS

The process of matching a named entity reference in a dataset to a unique, canonical identifier in a knowledge base—such as a Wikidata Q-ID or Google Knowledge Graph MID. This resolves ambiguity and consolidates authority signals by linking variant names to a single, authoritative node.

• Uses string similarity and contextual scoring
• Critical for deduplication in large catalogs
• Outputs a confidence score per match

A structured data model that represents entities as nodes and their relationships as edges, used by search engines to store deterministic facts and by enterprises to ground AI outputs in authoritative data. A controlled vocabulary defines the permissible node types and edge labels within this graph.

• Enables traversal queries and inference
• Powers Google's entity understanding
• Contrasts with unstructured vector embeddings