Basha

Basha is a lightweight, explainable and customizable Python library for sentiment and emotion analysis.

It analyses English text and returns:

• Sentiment label: positive, negative or neutral
• Continuous compound score from -1 to +1
• Positive, negative and neutral proportions
• Emotion information
• Sarcasm information
• Per-word explanations
• Social-media normalization information

Features

Basha currently supports:

• Weighted sentiment vocabulary
• Positive, negative and neutral classification
• Emotion detection
• Negation handling
• Intensifiers and diminishers
• Capital-letter emphasis
• Punctuation emphasis
• Contrast words such as but
• Emoji and emoticon sentiment
• Spelling correction for sentiment words
• Complete sentiment phrases
• Context-dependent words
• Basic sarcasm detection
• Social-media slang
• Hashtag conversion
• Username and URL removal
• Batch prediction
• Custom sentiment words
• Explainable output

Installation

After the package is published, install it using:

pip install basha-sentiment

For local development, open the terminal in the project folder and run:

python -m pip install --editable .

Basic usage

from basha import bashaanalyser


analyzer = bashaanalyser()

result = analyzer.predict(
    "This product is excellent"
)

print(result)

Get the sentiment label

from basha import bashaanalyser


analyzer = bashaanalyser()

result = analyzer.predict(
    "This product is excellent"
)

print(result["label"])

Output:

positive

Standard class-name usage

Python classes normally use capital letters. Basha also supports:

from basha import BashaAnalyser


analyzer = BashaAnalyser()

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

print(result["label"])

Example output

{
    "text": "This product is very good!",
    "label": "positive",
    "emotion": {
        "label": "joy",
        "confidence": 0.75
    },
    "scores": {
        "compound": 0.72,
        "positive": 0.25,
        "negative": 0.0,
        "neutral": 0.75
    },
    "flags": {
        "sarcasm": False,
        "normalized_text": "This product is very good!",
        "social_normalizations": []
    },
    "explanation": [
        {
            "token": "good",
            "base_score": 2.0,
            "final_score": 3.0,
            "rules": [
                "intensifier: very"
            ]
        }
    ]
}

The exact numerical values can change when the vocabulary and rules are updated.

Negation

print(
    analyzer.predict(
        "This product is not good"
    )["label"]
)

Output:

negative

Basha also handles positive negation:

print(
    analyzer.predict(
        "This product is not bad"
    )["label"]
)

Output:

positive

Intensifiers and diminishers

normal = analyzer.predict("good")
strong = analyzer.predict("very good")
weak = analyzer.predict("slightly good")

print(normal["scores"]["compound"])
print(strong["scores"]["compound"])
print(weak["scores"]["compound"])

very good should have a stronger positive score than good, while slightly good should have a weaker positive score.

Contrast handling

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

print(result["label"])

Output:

negative

The part appearing after but receives more importance.

Emoji and emoticon support

print(
    analyzer.predict(
        "I love this phone 😍"
    )["label"]
)

print(
    analyzer.predict(
        "This is bad :("
    )["label"]
)

Output:

positive
negative

Spelling correction

Basha can safely correct certain misspelled sentiment words:

result = analyzer.predict(
    "This product is graet"
)

print(result["label"])
print(result["explanation"])

The word graet can be interpreted as great.

Basha does not automatically correct every unknown word. It corrects only sufficiently clear spelling mistakes to reduce incorrect corrections.

Sarcasm detection

result = analyzer.predict(
    "Excellent service, yeah right."
)

print(result["label"])
print(result["flags"]["sarcasm"])
print(result["flags"]["sarcasm_reasons"])

Possible output:

negative
True
['sarcasm marker: yeah right']

Sarcasm detection is rule-based and cannot understand every form of sarcasm.

Social-media text

result = analyzer.predict(
    "@company this product is gr8! #AmazingProduct"
)

print(result["label"])
print(result["flags"]["normalized_text"])
print(result["flags"]["social_normalizations"])

Basha can:

• Convert gr8 to great
• Convert luv to love
• Split #AmazingProduct into amazing product
• Ignore usernames such as @company
• Ignore URLs
• Normalize repeated letters

Emotion detection

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

print(result["emotion"])

Basha can identify emotions supported by its emotion vocabulary, such as:

• Joy
• Sadness
• Anger
• Fear
• Surprise
• Disgust
• Neutral or uncertain emotion

Context-dependent words

Some words have different meanings in different situations:

print(
    analyzer.predict(
        "The movie was sick"
    )["label"]
)

print(
    analyzer.predict(
        "I feel sick"
    )["label"]
)

Possible output:

positive
negative

Other context-dependent words include words such as sharp, hot, sweet, fire and killer.

Custom words

You can add a word to one analyzer:

from basha import bashaanalyser


analyzer = bashaanalyser()

analyzer.add_word(
    "supercalifragilistic",
    3.5
)

result = analyzer.predict(
    "This is supercalifragilistic"
)

print(result["label"])

The custom score must be between -4.0 and +4.0.

Batch prediction

from basha import bashaanalyser


analyzer = bashaanalyser()

results = analyzer.predict_batch([
    "This product is excellent",
    "This product is terrible",
    "The parcel arrived today"
])

for result in results:
    print(result["label"])

Output:

positive
negative
neutral

Running tests

Install development tools:

python -m pip install --editable ".[dev]"

Run all tests:

python -m pytest

Passing tests prove that the tested behaviours work correctly. They do not prove complete real-world accuracy.

Benchmarking

Optional benchmark dependencies can be installed using:

python -m pip install --editable ".[benchmark]"

A benchmark can compare Basha with VADER and TextBlob using:

• Total execution time
• Average time per sentence
• Sentences processed per second

Do not claim that Basha is faster until benchmark results confirm it.

Project structure

basha/
├── src/
│   └── basha/
│       ├── __init__.py
│       ├── analyzer.py
│       ├── context.py
│       ├── emotions.py
│       ├── lexicon.py
│       ├── modifiers.py
│       ├── phrases.py
│       ├── sarcasm.py
│       ├── social.py
│       ├── spelling.py
│       ├── symbols.py
│       └── tokenizer.py
├── tests/
│   └── test_sentiment.py
├── example.py
├── LICENSE
├── pyproject.toml
└── README.md

Limitations

Basha is currently a rule-based English sentiment-analysis library.

Known limitations include:

• It does not learn automatically from new data
• Sarcasm detection is limited to recognized patterns
• Context understanding is limited to defined rules
• Spelling correction does not correct every unknown word
• Emotion detection depends on its emotion vocabulary
• Mixed sentiment may be simplified into one overall label
• Multilingual text is not yet fully supported
• Real-world accuracy must still be measured using labelled datasets

Development status

Basha is currently under active development. Its vocabulary, rules, output format and API may change before version 1.0.0.

Author

Bajil Mohammed E.P. Mohammed Shahbi

License

This project is licensed under the MIT License.