Skip to main content

Multimodal Contextual Precision

The multimodal contextual precision metric measures your RAG pipeline's retriever by evaluating whether nodes in your retrieval_context that are relevant to the given input are ranked higher than irrelevant ones. deepeval's multimodal contextual precision metric is a self-explaining MLLM-Eval, meaning it outputs a reason for its metric score.

info

The Multimodal Contextual Precision is the multimodal adaptation of DeepEval's contextual precision metric. It accepts images in addition to text for the input, retrieval_context, and expected_output.

Required Arguments​

To use the MultimodalContextualPrecisionMetric, you'll have to provide the following arguments when creating a MLLMTestCase:

  • input
  • actual_output
  • expected_output
  • retrieval_context

The input and actual_output are required to create an MLLMTestCase (and hence required by all metrics) even though they might not be used for metric calculation. Read the How Is It Calculated section below to learn more.

Example​

from deepeval import evaluate
from deepeval.metrics import MultimodalContextualPrecisionMetric
from deepeval.test_case import MLLMTestCase, MLLMImage

metric = MultimodalContextualPrecisionMetric()
test_case = MLLMTestCase(
    input=["Tell me about some landmarks in France"],
    actual_output=[
        "France is home to iconic landmarks like the Eiffel Tower in Paris.",
        MLLMImage(...)
    ],
    expected_output=[
        "The Eiffel Tower is located in Paris, France.",
        MLLMImage(...)
    ],
    retrieval_context=[
        MLLMImage(...),
        "The Eiffel Tower is a wrought-iron lattice tower built in the late 19th century.",
        MLLMImage(...)
    ],
)

metric.measure(test_case)
print(metric.score)
print(metric.reason)

# or evaluate test cases in bulk
evaluate([test_case], [metric])

There are SIX optional parameters when creating a MultimodalContextualPrecisionMetric:

  • [Optional] threshold: a float representing the minimum passing threshold, defaulted to 0.5.
  • [Optional] model: a string specifying which of OpenAI's Multimodal GPT models to use, OR any custom MLLM model of type DeepEvalBaseMLLM. Defaulted to 'gpt-4o'.
  • [Optional] include_reason: a boolean which when set to True, will include a reason for its evaluation score. Defaulted to True.
  • [Optional] strict_mode: a boolean which when set to True, enforces a binary metric score: 1 for perfection, 0 otherwise. It also overrides the current threshold and sets it to 1. Defaulted to False.
  • [Optional] async_mode: a boolean which when set to True, enables concurrent execution within the measure() method. Defaulted to True.
  • [Optional] verbose_mode: a boolean which when set to True, prints the intermediate steps used to calculate said metric to the console, as outlined in the How Is It Calculated section. Defaulted to False.

How Is It Calculated?​

The MultimodalContextualPrecisionMetric score is calculated according to the following equation:

Multimodal Contextual Precision=1Number of Relevant Nodes∑k=1n(Number of Relevant Nodes Up to Position kk×rk)\text{Multimodal Contextual Precision} = \frac{1}{\text{Number of Relevant Nodes}} \sum_{k=1}^{n} \left( \frac{\text{Number of Relevant Nodes Up to Position } k}{k} \times r_{k} \right)
info
  • k is the (i+1)th node in the retrieval_context
  • n is the length of the retrieval_context
  • rk is the binary relevance for the kth node in the retrieval_context. rk = 1 for nodes that are relevant, 0 if not.

The MultimodalContextualPrecisionMetric first uses an MLLM to determine for each node in the retrieval_context whether it is relevant to the input based on information in the expected_output, before calculating the weighted cumulative precision as the contextual precision score. The weighted cumulative precision (WCP) is used because it:

  • Emphasizes on Top Results: WCP places a stronger emphasis on the relevance of top-ranked results. This emphasis is important because MLLMs tend to give more attention to earlier nodes in the retrieval_context (which may cause downstream hallucination if nodes are ranked incorrectly).
  • Rewards Relevant Ordering: WCP can handle varying degrees of relevance (e.g., "highly relevant", "somewhat relevant", "not relevant"). This is in contrast to metrics like precision, which treats all retrieved nodes as equally important.

A higher multimodal contextual precision score represents a greater ability of the retrieval system to correctly rank relevant nodes higher in the retrieval_context.