Nasty.Operations.QA.CandidateScorer (Nasty v0.3.0)

Generic sentence scoring for Question Answering.

Scores sentences for relevance to a question using:

Keyword matching (lemma overlap)
Entity type matching (expected answer type)
Position bias (earlier sentences preferred)

Summary

Functions

score_sentence(sentence, question_analysis, all_sentences, config)

Scores a sentence for relevance to a question.

Functions

score_sentence(sentence, question_analysis, all_sentences, config)

@spec score_sentence(
  Nasty.AST.Sentence.t(),
  Nasty.Operations.QA.QuestionClassifier.t(),
  [{Nasty.AST.Sentence.t(), integer()}],
  map()
) :: float()

Scores a sentence for relevance to a question.

Returns a score between 0.0 and 1.0+, where higher scores indicate more relevant sentences.

Scoring Components

Base score: 0.1 (allows fallback even with no keyword matches)
Keyword match: 0.6 weight
Entity type: 0.2 weight
Position: 0.2 weight