← Blog

How to search the catalog by relevance, plurals and typos

  • Janium
  • search

Someone searching residencia and someone searching residencias should not get two different lists. Someone typing espanola without an accent should still find the heading with one. In keyword search in many catalogs each spelling is another query, and the order follows the date of entry: what was cataloged last comes first. The reader gets an empty list or one that does not start with what they were looking for.

In Janium 2027 that search can rank by relevance, recognize plurals and derived forms, and not distinguish accents when matching. If there are no exact hits, it can suggest the closest term. The version is in Janium 2027 is now available.

Rank by relevance

Relevance puts the most pertinent to the query first, not necessarily the last cataloged. It changes which record comes first. It is worth testing against the institution’s own holdings before leaving it as the default: an order that works for a book collection may not be the one expected by someone searching files.

If that order does not fit, the catalog is not stuck with it. Catalog order — the usual one, by record or by date — can be restored and relevance left off. Or the ranking scheme can be adjusted so some fields weigh more than others, according to how that holding is searched: a catalog searched mainly by title should not rank the same way as one searched by subject or by file.

Plurals, accents and derived forms

The engine recognizes plurals and derived forms. It ships with Spanish, English and Portuguese; other languages and other forms can be added when the catalog needs them. Residencia and residencias feed the same search. It also does not distinguish accents when matching: espanola finds the heading española. That does not replace authority control: the authorized form remains the one in the authority file; keyword search is a match over indexes.

When there is no exact hit

If there are no exact results, it can suggest the closest term: a typo or imperfect OCR. In the summary a snippet appears with the query words marked. While typing, suggestions come from the holdings themselves, not from a generic dictionary.

What does not go through this engine

The inverted index remains the default and the fallback. If the new engine is not turned on, keyword search behaves as it does today. Alphabetical browsing, authority control, ISBN, ISSN and classification do not go through that engine.

Collect and the catalog

Whoever uses Collect to catalog with AI still delivers records to Janium. 2027 changes how Janium searches by keyword, not how Collect produces the access points. Collect surfaces them once; Janium loads them into its indexes. A RAG does not replace that catalog. The relationship is in Access points, full text and RAG.

Limits

Relevance changes the order: it should be validated with catalogers and with how the public searches. If it does not fit, catalog order can be restored or the weights adjusted. When a language or a set of forms is added, the index is rebuilt so they enter the search. The closest-term suggestion appears when there is no exact match; it does not rewrite the query if there were already results. If the engine is left off, the search already in use does not change.

To continue the conversation

If you want to see how relevance behaves with your holdings —plurals, accents, typos— write to us at info@janium.com. How you search today changes whether it is worth leaving the engine as the default or trying it first.