AI Glossary / Retrieval and knowledge systems / Intermediate
Thesaurus (information retrieval)
Definition
A Thesaurus (information retrieval) is a structured vocabulary used to describe and retrieve information consistently. It connects preferred terms with alternative expressions and records relationships such as broader, narrower, and related concepts. Its purpose is to support indexing, browsing, and searching within a defined domain, rather than simply supplying interchangeable words for writing.
What it means
Authors and searchers often use different language for the same concept. A retrieval thesaurus helps connect those expressions, while distinguishing concepts that are related but not equivalent. A person searching with an accepted alternative label can be guided to the preferred concept used in the index.
The vocabulary also provides routes for exploration. A broader term can help someone widen a search, a narrower term can refine it, and a related term can suggest a different but relevant direction. These relationships should not be collapsed into one undifferentiated synonym list.
How it works
Define the domain and identify concepts that users need to find. Choose preferred labels, record alternative expressions, and add scope notes where labels alone are ambiguous. A stable concept identifier separates the concept from a display label that may change.
Establish hierarchical relationships for broader and narrower concepts and associative relationships for relevant connections outside that hierarchy. Some concepts may have more than one broader concept. Review relationships carefully because a convenient association does not necessarily justify a parent-child relationship.
Index resources using the concepts, then decide how the search application uses the thesaurus. It might map an alternative label to a concept, offer narrower choices, or expand a query through selected relationships. The vocabulary describes relationships; it does not by itself specify every ranking or expansion behavior.
SKOS is a standard model for representing concept schemes, including thesauri, with labels, notes, and relationships. A SKOS broader relationship should not simply be treated as an OWL subclass assertion. Applications also need to distinguish direct hierarchical links from traversal across several levels. Maintain vocabulary versions, deprecated concepts, and mappings so indexing remains understandable over time.
Three practical examples
- A specialized library uses preferred subject terms to connect publications that use different terminology.
- A business knowledge portal offers broader and narrower concepts to help users refine searches.
- A multilingual catalog attaches language-specific labels to shared concepts, with review of whether proposed translations express the intended meaning.
Think of it this way
Think of a signposted map of a subject. Alternative names point to the same destination, broader routes lead to larger areas, and related routes lead to neighboring destinations. Treating every sign as another name for the same place would make the map confusing.
Worked example
Consider an original, fictional thesaurus with the preferred concept label electric vehicle and the alternative label EV. Battery electric vehicle and plug-in hybrid vehicle are narrower concepts. Charging station is a related concept, not an alternative name for a vehicle.
Suppose resource A is indexed directly under electric vehicle. B and C are indexed under battery electric vehicle. D is indexed under plug-in hybrid vehicle. E is indexed under charging station. F is indexed under both electric vehicle and battery electric vehicle.
A query entered as EV first maps to electric vehicle. Under an exact-concept rule, it retrieves A and F: two resources. Under a stated rule that also includes the two narrower concepts, it retrieves A, B, C, D, and F: five distinct resources.
Counting individual matching assignments gives two direct assignments, three battery-electric assignments, and one plug-in-hybrid assignment, totaling six. The result set still contains five resources because F matches twice and must be deduplicated.
If the application additionally follows related-concept links, it adds E and returns six resources. That is a different expansion policy, not an automatic consequence of recognizing EV as an alternative label.
For a user asking specifically about vehicles, E may be an unwanted addition even though charging infrastructure is related. The example shows why equivalence, hierarchy, and association require different treatment. It does not establish the best policy for every query, and the invented vocabulary is not presented as an official classification.
Illustrative case study
Fictional case: Rowan Research combines reports from several business units. Reviewers find that teams use different labels for the same processes, while other similar-looking terms describe distinct activities.
They create a small retrieval thesaurus, add scope notes for ambiguous concepts, and preserve alternative labels for familiar departmental language. Search designers initially expand only through approved equivalences, then test optional narrower-concept expansion with users. Related concepts appear as suggestions rather than being silently treated as synonyms.
The team records vocabulary changes and revisits affected index assignments. This hypothetical case illustrates design, not a measured deployment.
Common misconceptions and mistakes
A retrieval thesaurus is not just a writing thesaurus. Its preferred terms, scope, and relationships are designed for information organization.
A related term is not a synonym, and a narrower concept is not interchangeable with its broader concept. Expanding every relationship can increase irrelevant results.
Controlled vocabulary does not eliminate ambiguity or cultural assumptions. Concepts and labels require review as domains and user language change. Automatic suggestions can help identify candidate relationships, but statistical similarity alone does not establish equivalence or a valid hierarchy.
Business use
Businesses can use retrieval thesauri to make content easier to find across departments, languages, and historical terminology. They are particularly useful when a stable subject structure must connect diverse wording without hiding important distinctions.
Start with recurring search failures and a manageable domain. Assign responsibility for concept definitions and changes, and test whether expansion improves relevant retrieval without excessive noise. Preserve identifiers and change records so old assignments can be interpreted or migrated. The maintenance effort should be evaluated alongside the benefits to users.
People also ask
Is a thesaurus the same as a taxonomy?
They overlap, but a thesaurus commonly emphasizes equivalence and associative relationships as well as hierarchy.
Is SKOS a thesaurus?
No. It is a model that can represent one.
Must every search expand to narrower terms?
No. Expansion is an application choice that should be explicit and evaluated.
Check your understanding
Why does EV retrieve the electric vehicle concept?
Answer: It is defined as an alternative label in the invented thesaurus.
Why are there five rather than six resources after hierarchical expansion?
Answer: F has two matching assignments but is one resource.
Why is charging station handled separately?
Answer: Its relationship is associative, not equivalence.
Related terms
- Controlled vocabulary
- Subject indexing
- preferred term
- alternative label
- broader term
- narrower term
- related term
- scope note
- SKOS
Learn next
Learn the three main relationship types, then construct a small vocabulary with stable concept identifiers and scope notes. Test exact matching, hierarchical expansion, and related-concept suggestions separately to understand their effects on retrieval.
More in Retrieval and knowledge systems
- Term-at-a-time - Term-at-a-time is a query-scoring strategy that processes one query term's postings before moving to another query term.
- Term discrimination
- Term partitioning - Term partitioning distributes an inverted index across machines by assigning different vocabulary terms to different partitions.
- Tiered indexes - Tiered indexes organize retrieval information into levels so a search can examine a promising subset first and fall back to broader levels when needed.
- Topic-specific PageRank - Topic-specific PageRank is a PageRank distribution biased toward a chosen topic by changing where the modeled random surfer teleports.
- Tragedy of the commons - The tragedy of the commons is a collective-action problem in which individually attractive use of a shared, limited resource can produce excessive use and a...
Sources
Primary standards guidance: W3C, SKOS Primer: https://www.w3.org/TR/skos-primer/ Primary normative specification: W3C, SKOS Reference: https://www.w3.org/TR/skos-reference/ Primary operational vocabulary example: NLM, Medical Subject Headings: https://www.nlm.nih.gov/mesh/meshhome.html
Previous: Theorem proving Next: Theta*