Sentix

A lightweight and explainable rule-based sentiment analysis library for Python.

Sentix provides sentiment classification with support for:

• Positive sentiment
• Negative sentiment
• Neutral sentiment
• Mixed sentiment
• Negation handling
• Intensity modifiers
• Diminishers
• Contrast detection
• Aspect-based sentiment
• Emotion detection
• Confidence scoring
• Evidence-based explanations

Installation

pip install sentix

Quick Start

from sentix import SentimentAnalyzer

analyzer = SentimentAnalyzer()

result = analyzer.predict(
    "I love the camera, but I hate the battery."
)

print(result.label)
print(result.score)
print(result.confidence)

Example output:

mixed
0.0
0.49

Full Result

from sentix import SentimentAnalyzer

analyzer = SentimentAnalyzer()

result = analyzer.predict(
    "The camera is amazing, but the battery is terrible."
)

print(result.to_dict())

Example structure:

{
    "label": "mixed",
    "score": 0.0,
    "positive": 0.5,
    "negative": 0.5,
    "neutral": 0.0,
    "confidence": 0.49,
    "emotions": {},
    "aspects": {},
    "evidence": []
}

Supported Sentiment Labels

Sentix can return four sentiment labels:

Label	Meaning
positive	Overall positive sentiment
negative	Overall negative sentiment
neutral	No significant sentiment detected
mixed	Both positive and negative sentiment detected

Negation

Sentix handles common negation patterns.

analyzer.predict("I don't like this.")

Expected sentiment:

negative

Example:

analyzer.predict("I don't think this is good.")

Expected sentiment:

negative

Intensity

Sentix supports intensity modifiers.

analyzer.predict(
    "This is extremely amazing."
)

The word extremely increases the strength of the sentiment.

Diminishers

Sentix also supports diminishing modifiers.

analyzer.predict(
    "The battery is slightly bad."
)

The negative sentiment is weaker than:

"The battery is terrible."

Contrast Handling

Sentix analyzes contrast expressions such as but.

analyzer.predict(
    "I love this, but the battery is terrible."
)

Contrast analysis considers sentiment before and after the contrast marker.

Mixed Sentiment

Sentix can detect conflicting sentiment.

analyzer.predict(
    "I love the screen, but I hate the battery."
)

Possible output:

mixed

Aspect-Based Sentiment

Sentix can identify sentiment associated with specific aspects.

result = analyzer.predict(
    "The camera is amazing but the battery is terrible."
)

print(result.aspects)

Example:

{
    "camera": {
        "sentiment": "positive",
        "score": 2.0
    },
    "battery": {
        "sentiment": "negative",
        "score": -2.0
    }
}

Emotion Detection

Sentix can detect emotion-related signals.

result = analyzer.predict(
    "I am extremely happy and excited."
)

print(result.emotions)

Evidence

Sentix provides explainable sentiment evidence.

result = analyzer.predict(
    "This product is extremely amazing!"
)

for item in result.evidence:
    print(item)

Evidence may include:

• sentiment word
• base score
• intensity modifier
• negation modifier
• capitalization modifier
• final score

Benchmark

Current benchmark results on the Sentix benchmark dataset:

Library	Accuracy	Macro F1
Sentix	97%	0.9632
VADER	83%	0.6793
TextBlob	69%	0.5264

Category Performance

Category	Sentix
Basic	100%
Negation	100%
Intensity	100%
Diminisher	100%
Contrast	80%
Mixed	90%
Aspect	100%
Morphology	100%
Emphasis	100%
Neutral	100%

Testing

Run the complete test suite:

python -m pytest

Current status:

173 passed

Benchmark

Run:

python benchmarks/benchmark_sentiment.py

Performance

On the current benchmark dataset of 100 samples:

Sentix: approximately 0.036 seconds

This is approximately:

0.36 milliseconds per sample

Performance will vary depending on hardware and Python version.

Design Philosophy

Sentix focuses on:

• Lightweight implementation
• Explainability
• Rule-based reasoning
• No external machine learning model requirement
• Fast inference
• Transparent sentiment evidence

License

MIT License

Author

Sentix Contributors