StemmingAlgorithm

sealed interface StemmingAlgorithm

How a full-text index reduces tokens to a stem.

It is not a tuning knob for every corpus. Stemming makes "running" match "run", which is what somebody searching prose wants and is wrong for a field of product codes or identifiers, and for a language whose stemmer is aggressive it loses distinctions the reader was relying on. Qdrant's default is no stemming, so this is a decision to opt into rather than one to undo.

Disabled exists because Qdrant 1.19 added it: before that the way to turn stemming off was a magic "none" language, which it now deprecates.

Inheritors

Types

Link copied to clipboard

No stemming, said explicitly, which overrides whatever the field's language would default to.

Link copied to clipboard
data class Snowball(val language: SnowballLanguage) : StemmingAlgorithm

The Snowball stemmer for one language.