feat(giskard-checks): add Sentiment check using textblob

[Koushik-Salammagari]

Description

Adds a new built-in Sentiment check that validates the sentiment polarity of model output text using the textblob library — no LLM API call required.

What's included

• Sentiment check (builtin/nlp_metrics.py) — registered as "sentiment" in the discriminated union registry
• 19 unit tests (tests/builtin/test_sentiment.py) — covering all acceptance criteria from the issue
• Optional dependency — textblob added under [project.optional-dependencies] nlp in pyproject.toml
• Exports — added to builtin/__init__.py and the top-level giskard.checks namespace
• uv.lock updated

Features

• Polarity score computed in [-1.0, 1.0] via TextBlob (runs locally, no network call)
• Sentiment label derived from polarity: "positive" (>0.05) / "negative" (<-0.05) / "neutral"
• expected: Literal["positive", "negative", "neutral"] | None — asserts the sentiment label
• min_score / max_score — asserts numeric polarity thresholds (both optional, each independently)
• text / text_key — supports direct values or JSONPath extraction from trace (default: trace.last.outputs)
• Polarity score attached as a Metric on every result
• Returns CheckStatus.ERROR with a helpful pip install 'giskard-checks[nlp]' message when textblob is not installed
• Full Pydantic round-trip serialisation/deserialisation support

Example usage

from giskard.checks import Sentiment, Scenario

# Assert the response is positive
scenario = (
    Scenario(name="positive_response")
    .interact(inputs="How is our service?", outputs="Your service is excellent!")
    .check(Sentiment(expected="positive"))
)

# Assert polarity is within a numeric range
check = Sentiment(
    text="This product is fantastic!",
    expected="positive",
    min_score=0.1,
)

# Just measure, no assertion (always passes — useful for monitoring)
check = Sentiment()

Install the optional dependency

pip install 'giskard-checks[nlp]'

Related Issue

Closes #2364

Type of Change

• 🚀 New feature (non-breaking change which adds functionality)

Checklist

• I've read the CODE_OF_CONDUCT.md document.
• I've read the CONTRIBUTING.md guide.
• I've written tests for all new methods and classes that I created.
• I've written the docstring in NumPy format for all the methods and classes that I created or modified.
• I've updated the uv.lock running uv lock

[gemini-code-assist]

Code Review

This pull request introduces a new Sentiment check to the giskard-checks library, leveraging the textblob library for local sentiment analysis without requiring LLM API calls. The implementation includes a new nlp_metrics.py module, integration into the package's built-in checks, and a comprehensive suite of unit tests. The nlp optional dependency group was also added to pyproject.toml. Review feedback focuses on improving type safety by using defined aliases, removing redundant type-ignore comments, and enhancing the clarity of success messages when score boundaries are partially specified.

[gemini-code-assist]

The return type of _polarity_to_label can be more specific by using the _SENTIMENT_LABELS type alias instead of a generic str.

Suggested change

	def _polarity_to_label(polarity: float) -> str:
	def _polarity_to_label(polarity: float) -> _SENTIMENT_LABELS:

[gemini-code-assist]

The type: ignore[valid-type] comment appears to be unnecessary here, as _SENTIMENT_LABELS is a valid Literal type alias. Removing it helps maintain a cleaner codebase.

Suggested change

	expected: _SENTIMENT_LABELS | None = Field( # type: ignore[valid-type]
	expected: _SENTIMENT_LABELS | None = Field(

[gemini-code-assist]

When either min_score or max_score is None, the success message currently displays None in the range (e.g., [0.1, None]). It would be clearer to use the actual polarity bounds (-1.0 and 1.0) as defaults in the message for better readability.

Suggested change

	if self.min_score is not None or self.max_score is not None:
	success_parts.append(
	f"Score {polarity:.4f} is within the required range "
	f"[{self.min_score}, {self.max_score}]."
	)
	if self.min_score is not None or self.max_score is not None:
	low = self.min_score if self.min_score is not None else -1.0
	high = self.max_score if self.max_score is not None else 1.0
	success_parts.append(
	f"Score {polarity:.4f} is within the required range [{low}, {high}]."
	)

[Koushik-Salammagari]

Hi team! This PR is ready for review — could a maintainer add the safe for build label so CI can run? Happy to address any feedback. Thanks!

[Koushik-Salammagari]

Hi! Just checking in on #2386 — the Sentiment check PR. All the Gemini bot suggestions have been addressed and the code is ready for review. Could a maintainer add the safe for build label so CI can run? Happy to address any feedback. Thanks!