Struct KeyBertAdapter 

pub struct KeyBertAdapter { /* private fields */ }

KeyBERT-backed Extractor adapter.

Construct with KeyBertAdapter::new; hand to crate::pipeline::Ingester::with_extractor in place of the default crate::extract::RuleExtractor.

Implementations

impl KeyBertAdapter

pub fn new(embedder: Arc<dyn Embedder>, label: impl Into<String>) -> Self

Build an adapter around the supplied embedder with KeyBERT defaults (top_k = 10, ngram_range = (1, 3), mmr_diversity = 0.5, pmi_threshold = 1.0).

label is the ntype string stamped on every entity this adapter emits. The caller owns the vocabulary — pass whatever label fits your graph (e.g. "Keyword", "Tag", "Concept", or any domain-specific type).

pub fn with_label(self, label: impl Into<String>) -> Self

Override the entity label. Returns self for chaining.

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

Override top_k. Returns self for chaining.

pub const fn with_pmi_threshold(self, t: f32) -> Self

Override the PMI threshold used when mining co-occurrence edges. Returns self for chaining.

Trait Implementations

impl Debug for KeyBertAdapter

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

Formats the value using the given formatter.

impl Extractor for KeyBertAdapter

fn prepare(&self, sections: &[Section]) -> Result<(), Error>

Optional pre-extraction hook. Called once per file by crate::pipeline::Ingester::ingest BEFORE any extract_entities / extract_relations call, with the full list of sections the file produced. The default implementation is a no-op, so existing extractors keep their behaviour.

fn extract_entities(&self, section: &Section) -> Vec<EntitySpan>

Extract entity mentions from a single section.

fn extract_relations( &self, entities: &[EntitySpan], section: &Section, ) -> Vec<RelationSpan>

Extract candidate relations between already-extracted entities.