Implementation:Explodinggradients Ragas AnswerCorrectness Metric

Field	Value
source	Repo
domains	Metrics, Evaluation
last_updated	2026-02-10

Overview

AnswerCorrectness measures the correctness of a generated answer compared to a ground truth reference by combining factual statement classification (TP/FP/FN) with optional semantic similarity scoring.

Description

The AnswerCorrectness class evaluates answer quality by decomposing both the response and reference into simplified statements, classifying each statement as True Positive, False Positive, or False Negative, and computing an F-beta score. This factuality score can be optionally combined with an embedding-based semantic similarity score using configurable weights (default 0.75 factuality, 0.25 similarity). The metric inherits from MetricWithLLM, MetricWithEmbeddings, and SingleTurnMetric.

Key attributes:

• weights -- A list of two floats controlling the balance between factuality and semantic similarity (default [0.75, 0.25]).
• beta -- Float controlling precision/recall tradeoff in the F-beta score (default 1.0).
• answer_similarity -- An optional AnswerSimilarity instance for computing semantic similarity.
• correctness_prompt -- The LLM prompt used for TP/FP/FN classification.
• statement_generator_prompt -- The LLM prompt used for decomposing text into statements.

Usage

The metric requires user_input, response, and reference columns in the input sample. An LLM must be configured, and embeddings must be set if the semantic similarity weight is non-zero.

Code Reference

Property	Value
Source Location	src/ragas/metrics/_answer_correctness.py L140-272
Class Signature	class AnswerCorrectness(MetricWithLLM, MetricWithEmbeddings, SingleTurnMetric)
Import	from ragas.metrics import AnswerCorrectness

I/O Contract

Inputs

Parameter	Type	Required	Description
user_input	str	Yes	The question or user query
response	str	Yes	The generated answer to evaluate
reference	str	Yes	The ground truth answer

Outputs

Output	Type	Description
score	float	Weighted combination of factuality F-beta score and semantic similarity (0.0 to 1.0)

Usage Examples

from ragas.metrics import AnswerCorrectness
from ragas.dataset_schema import SingleTurnSample

metric = AnswerCorrectness(weights=[0.75, 0.25], beta=1.0)
# metric.llm = ...  # Set your LLM
# metric.embeddings = ...  # Set your embeddings

sample = SingleTurnSample(
    user_input="What powers the sun?",
    response="The sun is powered by nuclear fusion.",
    reference="The sun is powered by nuclear fusion, where hydrogen atoms fuse to form helium."
)
# score = await metric.single_turn_ascore(sample)

A pre-configured instance is available:

from ragas.metrics._answer_correctness import answer_correctness

Related Pages

• Explodinggradients_Ragas_SemanticSimilarity_Metric -- Used internally for the semantic similarity component
• Explodinggradients_Ragas_Faithfulness_Metric -- Shares the StatementGeneratorPrompt for statement decomposition
• Explodinggradients_Ragas_FactualCorrectness_Metric -- Alternative factual evaluation metric using claim decomposition and NLI