A thesaurus describes hierarchy and association with broader, narrower, and related terms
Aliases: thesaurus · BT NT RT · broader narrower related
What it is
A thesaurus records three further relations among preferred terms: broader term, narrower term, and related term. It describes a concept network, not merely an alias list. BT/NT give a hierarchy that can be widened or narrowed; RT gives a cross-branch “see also.” Retrieval can walk these edges: too narrow, climb to a broader term; too wide, drop to a narrower one; off to the side, jump to a related term. Without the three edges, a controlled vocabulary only solves “one name per concept,” not “how to walk between concepts.”
A thesaurus hierarchy is a concept hierarchy. It need not be the site’s navigation tree. The two can be projected onto each other; they must not be assumed to be the same graph.
Why it happens
Queries often stop at the wrong grain. Someone who types “vehicle” may want “e-bike”; someone who types a model number may still need the series. BT/NT give the system a legal zoom, rather than guessing from substring inclusion. RT handles “not parent–child, but thought of together,” so relatedness is not stuffed in as a second narrower term that pollutes the hierarchy.
Relations must invert and not cycle: if B is narrower than A, A is broader than B; relatedness should be symmetric. Cycles and one-way stubs send automatic expansion the wrong way and pull in distant concepts. A thesaurus is worth its edges, not its term count.
Studying it
Test whether walking edges improves known-item and topical browse, and whether bad edges hurt precision.
- Paradigms: given an over-narrow or over-wide query, compare no edges, synonyms only, and expansion with BT/NT/RT; have indexers draw relations and measure agreement with a gold thesaurus. Check completeness and invertibility of relations against thesaurus-construction rules.
- Independent variables: which relation types are available, how many hops automatic expansion walks, whether RT is kept distinct from NT.
- Dependent variables: success on grain-mismatch tasks, irrelevant items introduced by expansion, whether people understand “this was widened to a parent concept.”
- Methodological note: treating the nav tree as BT/NT writes marketing categories into the concept network. Mark thesaurus relations at concept level, then decide which segment to project into the UI. Labs that only use synonym queries will not see what BT/NT uniquely do.
Where it stops holding
A flat vocabulary of a dozen exclusive states does not need thesaurus relations; forcing BT invents a fake hierarchy. Machine-learned similar terms are not RT: similarity is distributional closeness, RT is an editorial claim of association, and they fail differently. Multiple broader terms (one concept, two BTs) are legal but expansion policy must say which way to climb, or automatic widening forks. End-user UI need not display the abbreviations BT/NT/RT; the underlying relations still have to exist.
Applying it
- Fill BT/NT/RT on core concepts: every NT must invert to a BT, RTs come in pairs, and “related” must not be written as a child.
- Search “widen / narrow scope” should walk these edges and tell the person which broader or narrower term they landed on.
- If the nav tree projects thesaurus hierarchy, project only stable BT/NT segments. Do not write marketing buckets into the thesaurus.
- Verify with one over-wide and one over-narrow query on the same topic. The system should offer a legal drill-down or climb along the edges, and related terms must not appear as children in hierarchical controls.
Related
- Within the group: G1.10.1 A controlled vocabulary gives each concept one preferred term; others point to it · G1.10.3 Authority files hold canonical names for entities and resolve homographs and aliases · G1.10.4 Controlled vocabularies are staff-maintained, unlike folksonomies · G1.10.5 At retrieval, non-preferred terms map to the preferred term and yield the same results · G1.10.6 Vocabularies need versioning and change records as the domain evolves
- Adjacent: G1.03 Taxonomy and hierarchy · G1.09 Types of organizational structures · G3.16 Search scope and context
- Search terms:
thesaurus·broader term·related term
Cards in the same group
- G1.10.1A controlled vocabulary gives each concept one preferred term; others point to it
- G1.10.3Authority files hold canonical names for entities and resolve homographs and aliases
- G1.10.4Controlled vocabularies are staff-maintained, unlike folksonomies
- G1.10.5At retrieval, non-preferred terms map to the preferred term and yield the same results
- G1.10.6Vocabularies need versioning and change records as the domain evolves