Struct EntityExtractorService 

pub struct EntityExtractorService {
    llm: Option<Arc<dyn LlmProvider>>,
    domain: Domain,
    min_confidence: f32,
}

Service for extracting entities from text content.

Uses an LLM to identify named entities and their relationships, with graceful fallback when LLM is unavailable.

Fields

llm: Option<Arc<dyn LlmProvider>>

LLM provider for extraction.

domain: Domain

Default domain for extracted entities.

min_confidence: f32

Minimum confidence threshold for entities.

Implementations

impl EntityExtractorService

pub fn new(llm: Box<dyn LlmProvider>, domain: Domain) -> Self

Creates a new entity extractor with an LLM provider.

pub const fn without_llm(domain: Domain) -> Self

Creates an entity extractor without LLM (fallback mode only).

pub const fn with_shared_llm(llm: Arc<dyn LlmProvider>, domain: Domain) -> Self

Creates an entity extractor with a shared LLM provider.

pub const fn with_min_confidence(self, threshold: f32) -> Self

Sets the minimum confidence threshold for extracted entities.

pub fn extract(&self, text: &str) -> Result<ExtractionResult>

Extracts entities and relationships from text.

Arguments

• text - The text to extract entities from.

Returns

An ExtractionResult containing extracted entities and relationships.

Errors

Returns an error if LLM extraction fails and no fallback is possible.

fn extract_with_llm( &self, llm: &Arc<dyn LlmProvider>, text: &str, ) -> Result<ExtractionResult>

Extracts entities using LLM.

fn parse_llm_response(&self, response: &str) -> Result<LlmExtractionResponse>

Parses the LLM JSON response.

fn extract_json(&self, response: &str) -> String

Extracts JSON from a response that may be wrapped in markdown.

fn extract_json_from_markdown(&self, text: &str) -> Option<String>

Extracts JSON from a markdown code block.

fn extract_raw_json(&self, text: &str) -> Option<String>

Extracts a raw JSON object from text.

fn extract_fallback(&self, text: &str) -> ExtractionResult

Fallback extraction when LLM is unavailable.

Uses simple pattern matching for common entity patterns.

pub fn to_graph_entities(&self, extracted: &ExtractionResult) -> Vec<Entity>

Converts extracted entities to graph Entity objects.

Arguments

• extracted - The extraction result.

Returns

A vector of Entity objects ready for storage.

pub fn to_graph_relationships( &self, extracted: &ExtractionResult, entity_map: &HashMap<String, Entity>, ) -> Vec<Relationship>

Converts extracted relationships to graph Relationship objects.

Requires a mapping from entity names to entity IDs.

Arguments

• extracted - The extraction result.
• entity_map - Map from entity name to Entity.

Returns

A vector of Relationship objects ready for storage.

pub fn infer_relationships( &self, entities: &[Entity], ) -> Result<InferenceResult>

Infers relationships between existing entities.

Analyzes a set of entities and uses LLM to discover implicit relationships that weren’t explicitly stated in text.

Arguments

• entities - The entities to analyze for relationships.

Returns

An InferenceResult containing inferred relationships.

Errors

Returns an error if LLM inference fails and no fallback is possible.

fn infer_with_llm( &self, llm: &Arc<dyn LlmProvider>, entities: &[Entity], ) -> Result<InferenceResult>

Infers relationships using LLM.

fn format_entities_for_inference(&self, entities: &[Entity]) -> String

Formats entities for LLM inference.

fn parse_inference_response( &self, response: &str, ) -> Result<LlmInferenceResponse>

Parses the LLM JSON response for inference.

fn infer_fallback(&self, entities: &[Entity]) -> InferenceResult

Fallback inference when LLM is unavailable.

Uses heuristics to infer common relationships based on entity types.

pub fn inferred_to_graph_relationships( &self, inferred: &InferenceResult, entity_map: &HashMap<String, Entity>, ) -> Vec<Relationship>

Converts inferred relationships to graph Relationship objects.

Arguments

• inferred - The inference result.
• entity_map - Map from entity name to Entity.

Returns

A vector of Relationship objects ready for storage.