Overview
EmotionDetection is an output scanner that identifies emotional content in LLM responses by delegating to the input-side InputEmotionDetection scanner, with additional methods for detailed emotion analysis.
Description
The EmotionDetection output scanner is a thin wrapper around the corresponding input scanner InputEmotionDetection. It analyzes LLM outputs to detect the presence of specific emotions such as anger, fear, joy, sadness, surprise, and others. The scanner passes all keyword arguments directly to the underlying input scanner, providing full configurability. In addition to the standard scan method, this scanner offers two extra methods: get_emotion_analysis returns a dictionary mapping emotion labels to their confidence scores, and scan_with_full_output combines the standard scan result with the full emotion analysis dictionary in a single call. These additional methods make the scanner especially useful for analytics and monitoring use cases where detailed emotion breakdowns are needed.
Usage
Use this scanner when you need to monitor or control the emotional tone of LLM outputs. This is valuable for customer service chatbots (ensuring responses are empathetic and not angry), mental health applications (detecting distress signals), content moderation, and analytics dashboards that track the emotional profile of generated content over time.
Code Reference
Source Location
Signature
class EmotionDetection(Scanner):
def __init__(self, **kwargs) -> None: ...
def scan(self, prompt: str, output: str) -> tuple[str, bool, float]: ...
def get_emotion_analysis(self, output: str) -> dict[str, float]: ...
def scan_with_full_output(
self, prompt: str, output: str
) -> tuple[str, bool, float, dict[str, float]]: ...
Import
from llm_guard.output_scanners.emotion_detection import EmotionDetection
I/O Contract
Inputs
| Name |
Type |
Required |
Description
|
| prompt |
str |
Yes |
The input prompt
|
| output |
str |
Yes |
The LLM output to scan for emotional content
|
Constructor Parameters
| Name |
Type |
Required |
Default |
Description
|
| **kwargs |
Any |
No |
N/A |
All keyword arguments are passed through to InputEmotionDetection
|
Outputs
| Name |
Type |
Description
|
| sanitized_output |
str |
The output (potentially modified)
|
| is_valid |
bool |
Whether the output passed the scan
|
| risk_score |
float |
Risk score (-1.0 to 1.0)
|
Extended Outputs (scan_with_full_output)
| Name |
Type |
Description
|
| sanitized_output |
str |
The output (potentially modified)
|
| is_valid |
bool |
Whether the output passed the scan
|
| risk_score |
float |
Risk score (-1.0 to 1.0)
|
| emotion_analysis |
dict[str, float] |
Dictionary mapping emotion labels to confidence scores
|
Usage Examples
Basic Usage
from llm_guard.output_scanners.emotion_detection import EmotionDetection
scanner = EmotionDetection()
prompt = "How do I fix this error?"
output = "I understand how frustrating that must be. Let me help you resolve this issue."
sanitized_output, is_valid, risk_score = scanner.scan(prompt, output)
if not is_valid:
print(f"Concerning emotional content detected (risk: {risk_score})")
Detailed Emotion Analysis
from llm_guard.output_scanners.emotion_detection import EmotionDetection
scanner = EmotionDetection()
output = "I am absolutely thrilled to help you with this!"
emotions = scanner.get_emotion_analysis(output)
for emotion, score in sorted(emotions.items(), key=lambda x: x[1], reverse=True):
print(f" {emotion}: {score:.4f}")
Full Output Scan
from llm_guard.output_scanners.emotion_detection import EmotionDetection
scanner = EmotionDetection()
prompt = "Tell me about the weather"
output = "The weather is terrible and it makes everything miserable."
sanitized_output, is_valid, risk_score, emotions = scanner.scan_with_full_output(prompt, output)
print(f"Valid: {is_valid}, Risk: {risk_score}")
print(f"Emotions: {emotions}")
Related Pages
