estimate_confidence

estimate_confidence(mappings: Collection[SemanticMapping], *, confidence_model: Literal['binomial', 'mean'] | None = None, check: bool = True, precision: int | None = None) → float[source]

Estimate the confidence of a subject-predicate-triple based on multiple evidences.

Parameters:

  • mappings – A collection of mappings that all have the same subject-predicate-object triple. This algorithm explicitly handles when there is a negative predicate modifier.

  • confidence_model –

    Which confidence model to use when aggregating mapping confidences.

    • mean aggregation is \(\frac{1}{n} \sum_{i=1}^n c_i\)

    • binomial aggregation is \(1 - \prod_{i=1}^n (1 - c_i)\)

  • check – Should mappings be checked to all have the same subject-predicate-object triple? This can be disabled if you’re sure they already match

Returns:

A single floating point confidence estimate of the positive subject-predicate-object triple, where 1.0 is highly confident and 0.0 is not confident. To get the confidence for the negated subject-predicate-object triple, subtract this return value from 1.0.

Note

We define the confidence in an empty list to be 1.0