The Role of Thesaurus Designers in Information Retrieval

Every time you run a search in a scholarly database and get back exactly what you needed, someone made that possible long before you typed a single word. Behind academic databases, medical archives, and digital libraries sits a carefully built vocabulary that decides which words mean the same thing, which terms are preferred, and how concepts connect to one another. The professional who builds this vocabulary is the thesaurus designer. Their work is invisible to most users, yet it shapes whether a search returns precise, relevant results or buries you under noise. This post explains what thesaurus designers do, why they matter, and how their craft keeps modern information retrieval accurate.

What a thesaurus means in information retrieval 🔗

The word “thesaurus” probably reminds you of a writing tool that lists synonyms. In information science, it means something quite different. Here, a thesaurus is a controlled vocabulary designed to control how concepts are described during indexing and searching. It is a structured list of terms, the relationships between those terms, and a set of rules for using them.

The international standard ISO 25964 defines a thesaurus as a controlled and structured vocabulary where concepts are represented by terms, organised so the relationships between concepts are made explicit, and where preferred terms come with entries for synonyms. The key idea is consistency. When an indexer and a searcher both reach for the same preferred term to describe a subject, the search succeeds. When they pick different words for the same idea, relevant documents slip through the cracks.

A classic example shows the problem clearly. In a plain full-text search engine, a query for “automobile” will miss every document that uses the word “car” instead, because the system treats them as unrelated strings of letters. A thesaurus links these terms as equivalents, so the system can retrieve both. As one analysis of adaptive thesaurus systems explains, expanding a query with synonyms is one of the main ways to improve recall.

What the thesaurus designer actually builds 🔗

A thesaurus is not just a word list. It is a network of relationships, and building that network is the designer’s core job. Three relationship types form the backbone of every thesaurus.

Equivalence relationships 🔗

These link synonyms and near-synonyms to a single preferred term (also called a descriptor). The other words become non-preferred terms that point the user to the right entry. This is how the system ensures that “heart attack” and “myocardial infarction” lead to the same set of documents.

Hierarchical relationships 🔗

These connect broader terms (BT) and narrower terms (NT), creating a tree of concepts from general to specific. A search for a broad topic can then expand to include its narrower children. Modern standards even distinguish different kinds of hierarchy, such as generic, partitive (whole-part), and instance relationships, as documented in the ISO 25964 data model.

Associative relationships 🔗

These link related terms (RT) that are not synonyms and not part of the same hierarchy but share a meaningful connection. They act as extra access points, guiding users toward related concepts they might not have thought to search for and widening the path to relevant material.

How a thesaurus designer approaches the work 🔗

Creating a thesaurus is a structured, iterative process that blends domain knowledge with information science principles. The designer rarely works alone and rarely gets it right on the first attempt.

Understanding the domain 🔗

The work begins with deep immersion in the subject field. Designers read research papers, textbooks, glossaries, and existing vocabularies to learn how specialists actually use terminology. They consult subject matter experts to catch the subtle distinctions that outsiders miss. A term that looks straightforward can carry different meanings across two sub-fields, and the designer has to resolve that ambiguity.

Collecting and selecting terms 🔗

Designers gather candidate terms from the documents the system will index and from the language real users type into searches. There are broadly two methods. One is building the vocabulary from scratch by analysing a document collection. The other, useful when good vocabularies already exist, is merging two or more thesauri into one, a strategy discussed in the classic literature on thesaurus construction. The challenge in merging is to combine sources without breaking the internal logic of any one of them.

Organising with facet analysis 🔗

To impose order on hundreds or thousands of terms, designers often use facet analysis, grouping concepts into fundamental categories before arranging hierarchies. ISO 25964 explicitly covers facet analysis, node labels, and presentation as part of good thesaurus construction. This structure is what lets a database offer clean browsing by category instead of a flat, confusing list.

Testing and maintaining 🔗

Design is never finished at launch. As new documents arrive and user feedback comes in, the vocabulary needs revision. For this reason it helps to separate the role of designer from that of long-term maintainer, as noted in the Library of Congress training materials on controlled vocabulary and thesaurus design. A thesaurus that is not updated slowly drifts out of step with its field and loses its usefulness.

Why thesaurus design matters for search quality 🔗

The payoff of all this work is measured in two metrics that every information retrieval system is judged by: precision and recall. As explained through worked examples, precision is the share of returned results that are actually relevant, while recall is the share of all relevant documents that the search manages to retrieve. The two are usually in tension; pushing one up often drags the other down.

A well-built thesaurus improves both sides of this trade-off at once. By linking synonyms, it raises recall, because a single query can pull in documents that used different wording. By controlling ambiguity and pinning each concept to one preferred term, it raises precision, because the system stops returning documents that merely share a stray keyword. This balance is exactly what general web searching often lacks and what specialised databases get right.

Thesauri at work in the real world 🔗

The clearest way to see a thesaurus designer’s value is to look at the systems their work powers.

Medical Subject Headings (MeSH) 🔗

The most famous example in the health sciences is MeSH, the controlled vocabulary thesaurus created and maintained by the US National Library of Medicine and used to index PubMed and MEDLINE. According to a university library guide on controlled vocabulary, MeSH organises over 24,000 descriptors into a hierarchical tree structure. When PubMed assigns a single MeSH heading to cover bacterial counts, fungal counts, and spore counts, a searcher using that heading finds all of them at once, without guessing every variant term. Medical researchers rely on this so heavily that, among major platforms, only PubMed indexes its literature with a standardised vocabulary like MeSH.

AGROVOC for agriculture 🔗

In agriculture, the Food and Agriculture Organization of the United Nations maintains AGROVOC, a multilingual thesaurus widely used in specialised and digital libraries to index content and support text mining. It has grown into a structured collection of more than 41,000 concepts expressed through nearly a million terms across dozens of languages, as described by the Interoperable Europe portal. AGROVOC powers FAO’s own bibliographic databases, showing how a single well-designed vocabulary can serve an entire global discipline.

These examples matter directly for students and researchers in India, where agricultural research institutions, medical libraries, and university databases depend on exactly this kind of structured vocabulary to make their collections searchable.

The thesaurus designer in the digital age 🔗

The role has not stood still. Today’s designers increasingly turn traditional thesauri into machine-readable knowledge organisation systems for the web. The key technology here is SKOS (Simple Knowledge Organization System), a W3C recommendation for publishing vocabularies on the semantic web. ISO 25964-1 advises on how to select and fit together concepts, terms, and relationships, while SKOS handles porting the finished thesaurus to the web in a format machines can reason over.

AGROVOC again illustrates the shift: it has been published as linked open data and aligned with other multilingual vocabularies, turning a once-printed index into a connected node in a global web of knowledge. For thesaurus designers, this means new skills in linked data, ontology development, and interoperability, sitting alongside the traditional craft of choosing the right preferred term. The job is evolving from compiling word lists toward engineering the semantic infrastructure of digital information.

What endures, though, is the core purpose. Whether on paper in the 1980s or in a linked data cloud today, the thesaurus designer builds the bridge between the words a searcher uses and the concepts an archive contains. That bridge is what turns a chaotic pile of documents into a system you can actually search.

What do you think? Have you noticed how searching a specialised academic or medical database feels different from a general web search, and which one tends to give you cleaner results? If you were designing a thesaurus for your own field of study, which terms do you think would cause the most confusion and need the most careful control?

References

1. https://en.wikipedia.org/wiki/Controlled_vocabulary
2. https://www.niso.org/schemas/iso25964
3. https://image-ppubs.uspto.gov/dirsearch-public/print/downloadPdf/8463806
4. https://link.springer.com/article/10.1007/s00799-015-0162-2
5. http://orion.lcg.ufrj.br/Dr.Dobbs/books/book5/chap09.htm
6. https://www.loc.gov/catworkshop/courses/thesaurus/pdf/cont-vocab-thes-instr-manual.pdf
7. https://opensourceconnections.com/blog/2016/03/30/search-precision-and-recall-by-example/
8. https://guides.library.illinois.edu/c.php?g=1191165&p=8712542
9. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4878086/
10. https://aims.fao.org/news/use-agrovoc-thesaurus-common-vocabulary-cgiar-core-metadata-schema-and-application
11. https://interoperable-europe.ec.europa.eu/collection/eu-semantic-interoperability-catalogue/solution/agrovoc-thesaurus
12. https://www.niso.org/standards-committees/iso-25964