Struct SemanticDiversifier 

pub struct SemanticDiversifier {
    pub config: DiversifierConfig,
    /* private fields */
}

Diversifies a ranked list of candidates using Maximal Marginal Relevance.

Each call to select greedily picks items that simultaneously maximise relevance to the query and minimise similarity to previously selected items.

Fields

config: DiversifierConfig

MMR configuration (lambda, etc.).

Implementations

impl SemanticDiversifier

pub fn new(config: DiversifierConfig) -> Self

Create a new diversifier with the given configuration.

pub fn select( &mut self, _query: &[f32], candidates: Vec<DiversificationCandidate>, k: usize, ) -> Vec<DiversifiedResult>

Select up to k diversified results from candidates using MMR.

The algorithm greedily builds a selected set S, starting from empty. Each iteration scores every remaining candidate with:

mmr = lambda * relevance_score - (1 - lambda) * max_sim_to_selected

Ties are broken by doc_id ascending.

Returns an empty Vec when candidates is empty or k == 0.

pub fn stats(&self) -> &DiversifierStats

Return a reference to the accumulated statistics.

pub fn set_lambda(&mut self, lambda: f32)

Override the lambda trade-off parameter, clamped to [0.0, 1.0].