Struct KeyBertExtractor 

pub struct KeyBertExtractor<'a> {
    pub embedder: &'a dyn Embedder,
    pub top_k: usize,
    pub ngram_range: (usize, usize),
    pub mmr_diversity: f32,
}

KeyBERT-style extractor.

Holds a borrowed Embedder reference; the caller owns the concrete provider (Ollama / OpenAI / ONNX / mock) and threads it in for the duration of an ingest run.

Fields

embedder: &'a dyn Embedder

Embedder used to encode candidate n-grams. MUST be the same provider + model that produced chunk_embed, otherwise cosine similarity is meaningless.

top_k: usize

Number of entities to return per call. See DEFAULT_TOP_K.

ngram_range: (usize, usize)

Inclusive (min_n, max_n) n-gram length range.

mmr_diversity: f32

MMR diversity coefficient in [0.0, 1.0].

Implementations

impl<'a> KeyBertExtractor<'a>

pub fn new(embedder: &'a dyn Embedder) -> Self

Construct a KeyBERT extractor with default parameters.

pub const fn with_top_k(self, k: usize) -> Self

Override top_k. Returns self for chaining.

pub const fn with_ngram_range(self, min: usize, max: usize) -> Self

Override n-gram range. min must be >= 1; callers that pass 0 are clamped to 1. Returns self for chaining.

pub fn with_mmr_diversity(self, lambda: f32) -> Self

Override MMR diversity coefficient. Returns self for chaining. Callers passing values outside [0.0, 1.0] get them clamped.

Trait Implementations

impl Debug for KeyBertExtractor<'_>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Extractor for KeyBertExtractor<'_>

fn extract_entities(&self, text: &str, chunk_embed: &[f32]) -> Vec<Entity>

Extract entity mentions from text.

fn extract_relations(&self, text: &str, entities: &[Entity]) -> Vec<Relation>

Mine candidate relations over an already-extracted entity set.

fn infer_typed_relations( &self, _text: &str, _entities: &[Entity], _budget: &InferenceBudget, ) -> Vec<TypedRelation>

Optionally infer typed relations between already-extracted entities, subject to the supplied [InferenceBudget].