Implementation:Protectai Llm guard Input Sentiment

Knowledge Sources	Protectai_Llm_guard
Domains	Sentiment_Analysis, NLP
Last Updated	2026-02-14 12:00 GMT

Overview

The Sentiment scanner detects negative sentiment in prompts using NLTK's VADER (Valence Aware Dictionary and sEntiment Reasoner) lexicon-based analyzer.

Description

Sentiment is an input scanner that evaluates the overall sentiment of a prompt using NLTK's SentimentIntensityAnalyzer with the VADER lexicon. Unlike other scanners in LLM Guard that use HuggingFace transformer models, this scanner relies on a rule-based, lexicon-driven approach that is extremely fast and requires no GPU. VADER produces a compound sentiment score ranging from -1.0 (most negative) to +1.0 (most positive). The threshold parameter (default -0.3) defines the cutoff: prompts with a compound score below the threshold are flagged as having negative sentiment. The lexicon parameter allows specifying an alternative VADER lexicon file if needed.

Usage

Use the Sentiment scanner for lightweight, fast sentiment filtering when you need to detect overtly negative or hostile prompts. It is best suited for applications where speed is critical and where coarse-grained sentiment detection is sufficient. For more nuanced emotional analysis, consider the EmotionDetection scanner instead.

Code Reference

Source Location

• Repository: Protectai_Llm_guard
• File: llm_guard/input_scanners/sentiment.py
• Lines: 1-59

Signature

class Sentiment(Scanner):
    def __init__(
        self,
        *,
        threshold: float = -0.3,
        lexicon: str = "vader_lexicon",
    ) -> None: ...

    def scan(self, prompt: str) -> tuple[str, bool, float]: ...

Import

from llm_guard.input_scanners import Sentiment

I/O Contract

Inputs

Name	Type	Required	Description
threshold	float	No	Compound sentiment score below which a prompt is flagged as negative. Range: -1.0 to 1.0. Defaults to -0.3.
lexicon	str	No	The NLTK VADER lexicon to use. Defaults to "vader_lexicon".

scan() Inputs

Name	Type	Required	Description
prompt	str	Yes	The input text to analyze for sentiment.

Outputs

Name	Type	Description
prompt	str	The original prompt (unchanged).
is_valid	bool	True if the compound sentiment score is at or above the threshold; False if below.
risk_score	float	The absolute value of the compound sentiment score, normalized between 0.0 and 1.0.

Sentiment Score Interpretation

VADER produces a compound score with the following general interpretation:

Score Range	Interpretation
+0.05 to +1.0	Positive sentiment
-0.05 to +0.05	Neutral sentiment
-1.0 to -0.05	Negative sentiment

The default threshold of -0.3 is calibrated to flag moderately to strongly negative prompts while allowing mildly negative or neutral text through.

Usage Examples

Basic Usage

from llm_guard.input_scanners import Sentiment

scanner = Sentiment()
prompt = "This is absolutely terrible and I hate everything about it!"
sanitized_prompt, is_valid, risk_score = scanner.scan(prompt)

print(is_valid)    # False (negative sentiment below -0.3)
print(risk_score)  # High risk score

Custom Threshold

from llm_guard.input_scanners import Sentiment

# Stricter threshold - flags mildly negative text too
scanner = Sentiment(threshold=0.0)
prompt = "I'm not sure this is a good idea"
sanitized_prompt, is_valid, risk_score = scanner.scan(prompt)

print(is_valid)    # May be False depending on VADER score
print(risk_score)

Positive Sentiment Check

from llm_guard.input_scanners import Sentiment

scanner = Sentiment(threshold=-0.3)
prompt = "Thank you so much for the wonderful help!"
sanitized_prompt, is_valid, risk_score = scanner.scan(prompt)

print(is_valid)    # True (positive sentiment)
print(risk_score)  # Low risk score

Related Pages

• Principle:Protectai_Llm_guard_Sentiment_Analysis