Struct KeywordMatcher 

pub struct KeywordMatcher {
    pub name_weight: usize,
    pub type_weight: usize,
    pub desc_weight: usize,
    pub synonyms: HashMap<String, Vec<String>>,
    pub use_tfidf: bool,
    pub synonym_weight: f64,
    pub cjk_bigram_weight: f64,
}

Matches entities by keyword substring search.

This is the default matcher used by GraphRAG. It splits the query into keywords and scores each entity based on how many keywords match the entity’s name, type, and description. Name matches are weighted highest.

P2-4: on top of the fixed name+3/type+2/desc+1 weights, three improvements fix the arbitrary weights and the recall gaps of substring matching for synonyms, polysemes, and Chinese-English mixed text:

• Synonym-table expansion synonyms: when a query term hits a synonym key, the equivalent words are matched too (each hit decayed by synonym_weight, default 0.7).
• Chinese-English mixed normalization: full-width -> half-width plus splitting long Chinese queries into CJK bigrams, fixing the problem that a space-free Chinese single token cannot hit a short entity name.
• TF-IDF weighting: each query term is weighted by its inverse document frequency in the entity corpus; common words (e.g. “Technology”) discriminate little and contribute little, while rare words contribute more; use_tfidf can disable it.

Fields

name_weight: usize

Weight for name matches (default: 3).

type_weight: usize

Weight for type matches (default: 2).

desc_weight: usize

Weight for description matches (default: 1).

synonyms: HashMap<String, Vec<String>>

Synonym table: query term (normalized lowercase/half-width form) -> list of equivalent words, also in normalized form. When an equivalent word is hit, it is matched once more with the contribution multiplied by synonym_weight.

use_tfidf: bool

Whether TF-IDF weighting is enabled (default true). When disabled, falls back to the fixed weights.

synonym_weight: f64

Decay factor for synonym hits (default 0.7).

cjk_bigram_weight: f64

Decay factor for CJK bigram hits (default 0.5).

Implementations

impl KeywordMatcher

pub fn new() -> Self

Creates a new keyword matcher with default weights.

pub fn with_synonyms(self, synonyms: HashMap<String, Vec<String>>) -> Self

Configures the synonym table (query term -> equivalent word list).

pub fn with_tfidf(self, enabled: bool) -> Self

Toggles TF-IDF weighting (enabled by default).

Trait Implementations

impl Default for KeywordMatcher

fn default() -> Self

Returns the “default value” for a type.

impl EntityMatcher for KeywordMatcher

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.