Struct rust_bert::pipelines::keywords_extraction::KeywordExtractionConfig

pub struct KeywordExtractionConfig<'a> {
    pub sentence_embeddings_config: SentenceEmbeddingsConfig,
    pub tokenizer_stopwords: Option<HashSet<&'a str>>,
    pub tokenizer_pattern: Option<Regex>,
    pub scorer_type: KeywordScorerType,
    pub ngram_range: (usize, usize),
    pub num_keywords: usize,
    pub diversity: Option<f64>,
    pub max_sum_candidates: Option<usize>,
}

Configuration for Keyword extraction

Fields

sentence_embeddings_config: SentenceEmbeddingsConfig

SentenceEmbeddingsConfig defining the sentence embeddings model to use

tokenizer_stopwords: Option<HashSet<&'a str>>

Optional list of tokenizer stopwords to exclude from the keywords candidate list. Default to a list of English stopwords.

tokenizer_pattern: Option<Regex>

Optional tokenization regex pattern. Defaults to sequence of word characters.

scorer_type: KeywordScorerType

KeywordScorerType used to rank keywords.

ngram_range: (usize, usize)

N-gram range (inclusive) for keywords. (1, 2) would consider all 1 and 2 word gram for keyword candidates.

num_keywords: usize

Number of keywords to return

diversity: Option<f64>

Optional diversity parameter used for the MaximalMarginRelevance ranker, defaults to 0.5. A high diversity (closer to 1.0) will give more importance to getting varied keywords, at the cost of less relevance to the original document.

max_sum_candidates: Option<usize>

Optional number of candidate sets used for MaxSum ranker. Higher values are more likely to identify a global optimum for the ranker criterion, but are more likely to include sets that are less relevant to the input document. Larger values also have a higher computational and memory cost (N² scale)

Trait Implementations

impl Default for KeywordExtractionConfig<'_>

fn default() -> Self

Returns the “default value” for a type.