Struct RelationExtractor 

pub struct RelationExtractor { /* private fields */ }

Relation extraction module

Implementations

impl RelationExtractor

pub fn new(_config: &AiConfig) -> Result<Self>

Create a new relation extractor backed by the built-in heuristic (rule-based) NER, relation classifier, and entity linker.

These heuristic backends are transparent, deterministic, and honest: they do NOT emulate a trained ML model, they do not fabricate confidence scores dressed up as model probabilities, and the entity linker does not invent unverified knowledge-base URIs. Every relation they produce is tagged with metadata["extraction_method"] = "heuristic-keyword-match" so downstream consumers can distinguish heuristic output from a real model’s.

To use a real ML backend (BERT-NER, a trained relation classifier, a live entity-linking service, etc.), construct real trait objects and pass them to with_backends.

_config is currently unused by the heuristic backends (they have no tunable model parameters); it is retained for API symmetry with the backend-injecting constructor.

pub fn with_backends( extraction_config: ExtractionConfig, ner_model: Box<dyn NamedEntityRecognizer>, relation_model: Box<dyn RelationClassifier>, entity_linker: Box<dyn EntityLinker>, ) -> Self

Create a relation extractor with caller-provided backends.

This is the path for wiring in a real NER model, relation classifier, and entity linker. The extraction_config’s confidence_threshold is used to filter extracted relations.

pub async fn extract_relations( &self, text: &str, ) -> Result<Vec<ExtractedRelation>>

Extract relations from text

pub fn to_triples(&self, relations: &[ExtractedRelation]) -> Result<Vec<Triple>>

Convert extracted relations to RDF triples