Struct EmbeddingMatcher 

pub struct EmbeddingMatcher<E: Embeddings> { /* private fields */ }

Matches entities by computing embedding similarity between the query and entity representations (name + type + description).

Requires an Embeddings implementation to compute vectors. Embeddings are cached internally to avoid recomputation across calls.

Implementations

impl<E: Embeddings> EmbeddingMatcher<E>

pub fn new(embeddings: E) -> Self

Creates a new embedding matcher with the given embeddings backend.

pub fn with_min_score(self, min_score: f64) -> Self

Sets the minimum embedding-similarity threshold (P1-2), default 0.0 keeps the old behavior.

impl<E: Embeddings + 'static> EmbeddingMatcher<E>

pub async fn find_relevant_async( &self, query: &str, store: &GraphStore, top_k: usize, ) -> Result<Vec<String>, GraphRAGError>

Async version of entity matching using embeddings.

This is the preferred method when using embedding-based matching, since embedding computation is inherently async.

P0-2: no more silent degradation / silent 0 scores — embedding failures or vector dimension mismatches now error out explicitly, letting callers know semantic matching is unavailable or the data is defective, instead of quietly falling back to keyword or treating “dimension mismatch” as “dissimilar”.

Trait Implementations

impl<E: Embeddings + 'static> EntityMatcher for EmbeddingMatcher<E>

fn find_relevant( &self, query: &str, _store: &GraphStore, _top_k: usize, ) -> Vec<String>

Find entity IDs relevant to the query, returning at most top_k results.