visonorm 0.1.3

Vietnamese Social Media Lexical Normalization Toolkit

📦 ViSoNorm Toolkit — Vietnamese Text Normalization & Processing

ViSoNorm is a specialized toolkit for Vietnamese text normalization and processing, optimized for NLP environments and easily installable via PyPI. Resources (datasets, models) are stored and managed directly on Hugging Face Hub and GitHub Releases.

---

🚀 Key Features

1. 🔧 BasicNormalizer — Basic Text Normalization

• Case folding: convert entire text to lowercase/uppercase/capitalize.
• Tone normalization: normalize Vietnamese tone marks.
• Basic preprocessing: remove extra whitespace, special characters, sentence formatting.

2. 😀 EmojiHandler — Emoji Processing

• Detect emojis: detect emojis in text.
• Split emoji text: separate emojis from sentences.
• Remove emojis: remove all emojis.

3. ✏️ Lexical Normalization — Social Media Text Normalization

• ViSoLexNormalizer: Normalize text using deep learning models from HuggingFace.
• NswDetector: Detect non-standard words (NSW).
• detect_nsw(): Utility function to detect NSW.
• normalize_sentence(): Utility function to normalize sentences.

4. 📊 Resource Management — Dataset Management

• list_datasets() — List available datasets.
• load_dataset() — Load dataset from GitHub Releases.
• get_dataset_info() — View detailed dataset information.

5. 🧠 Task Models — Task Processing Models

• SpamReviewDetection — Spam detection.
• HateSpeechDetection — Hate speech detection.
• HateSpeechSpanDetection — Hate speech span detection.
• EmotionRecognition — Emotion recognition.
• AspectSentimentAnalysis — Aspect-based sentiment analysis.

---

📥 Installation

Install from PyPI (Recommended)

pip install visonorm

Requirements

• Python >= 3.10
• PyTorch >= 1.10.0
• Transformers >= 4.0.0
• scikit-learn >= 0.24.0
• pandas >= 1.3.0

---

📚 Usage Guide

1. 🔧 BasicNormalizer — Basic Text Normalization

from visonorm import BasicNormalizer

# Initialize BasicNormalizer
normalizer = BasicNormalizer()

# Example text
text = "Hôm nay tôi rất VUI 😊 và HẠNH PHÚC 🎉!"

# Case folding
print(normalizer.case_folding(text, mode='lower'))
# Output: hôm nay tôi rất vui 😊 và hạnh phúc 🎉!

print(normalizer.case_folding(text, mode='upper'))
# Output: HÔM NAY TÔI RẤT VUI 😊 VÀ HẠNH PHÚC 🎉!

print(normalizer.case_folding(text, mode='capitalize'))
# Output: Hôm Nay Tôi Rất Vui 😊 Và Hạnh Phúc 🎉!

# Tone normalization
text2 = "Bận xong rồi. Xoã đi :)"
print(normalizer.tone_normalization(text2))
# Output: Bận xong rồi. Xõa đi :)

# Basic normalization with options
normalized = normalizer.basic_normalizer(
    text,
    case_folding=True,
    mode='lower',
    remove_emoji=False,
    split_emoji=True
)
print(normalized)
# Output: ['hôm', 'nay', 'tôi', 'rất', 'vui', '😊', 'và', 'hạnh', 'phúc', '🎉', '!']

# Remove emojis
normalized_no_emoji = normalizer.basic_normalizer(
    text,
    case_folding=True,
    remove_emoji=True
)
print(normalized_no_emoji)
# Output: ['hôm', 'nay', 'tôi', 'rất', 'vui', 'và', 'hạnh', 'phúc', '!']

2. 😊 EmojiHandler — Emoji Processing

from visonorm import EmojiHandler

# Initialize EmojiHandler
emoji_handler = EmojiHandler()

text = "Hôm nay tôi rất vui 😊🎉😊 và hạnh phúc 🎉!"

# Detect emojis
emojis = emoji_handler.detect_emoji(text)
print(f"Detected emojis: {emojis}")
# Output: Detected emojis: ['😊🎉😊', '🎉']

# Split emoji text
split_text = emoji_handler.split_emoji_text(text)
print(f"Split emoji text: {split_text}")
# Output: Hôm nay tôi rất vui 😊 🎉 😊 và hạnh phúc 🎉 !

# Split consecutive emojis
text_consecutive = "Hôm nay tôi rất vui 😊🎉😊"
split_consecutive = emoji_handler.split_emoji_emoji(text_consecutive)
print(f"Split consecutive: {split_consecutive}")
# Output: Hôm nay tôi rất vui 😊 🎉 😊

# Remove emojis
text_no_emoji = emoji_handler.remove_emojis(text)
print(f"Text without emojis: {text_no_emoji}")
# Output: Hôm nay tôi rất vui và hạnh phúc !

3. ✏️ Lexical Normalization — Social Media Text Normalization

Using ViSoLexNormalizer

from visonorm import ViSoLexNormalizer

# Initialize with default model (visolex/visobert-normalizer-mix100)
normalizer = ViSoLexNormalizer()

# Or specify a specific model from HuggingFace
# normalizer = ViSoLexNormalizer(model_repo="visolex/visobert-normalizer-mix100")
# normalizer = ViSoLexNormalizer(model_repo="visolex/bartpho-normalizer-mix100")

# Normalize sentence
input_str = "sv dh gia dinh chua cho di lam :))"
normalized = normalizer.normalize_sentence(input_str)
print(f"Original: {input_str}")
print(f"Normalized: {normalized}")
# Output:
# Original: sv dh gia dinh chua cho di lam :))
# Normalized: sinh viên đại học gia đình chưa cho đi làm :))

# Normalize and detect NSW simultaneously
nsw_spans, normalized_text = normalizer.normalize_sentence(input_str, detect_nsw=True)
print(f"Normalized: {normalized_text}")
print("Detected NSW:")
for nsw in nsw_spans:
    print(f"  - '{nsw['nsw']}' → '{nsw['prediction']}' (confidence: {nsw['confidence_score']})")
# Output:
# Normalized: sinh viên đại học gia đình chưa cho đi làm :))
# Detected NSW:
#   - 'sv' → 'sinh viên' (confidence: 1.0)
#   - 'dh' → 'đại học' (confidence: 1.0)
#   - 'dinh' → 'đình' (confidence: 1.0)
#   - 'chua' → 'chưa' (confidence: 1.0)
#   - 'di' → 'đi' (confidence: 1.0)
#   - 'lam' → 'làm' (confidence: 1.0)

Using NswDetector

from visonorm import NswDetector

# Initialize detector
detector = NswDetector()

# Detect NSW
input_str = "sv dh gia dinh chua cho di lam"
nsw_spans = detector.detect_nsw(input_str)
for nsw in nsw_spans:
    print(f"NSW: '{nsw['nsw']}' → '{nsw['prediction']}' (confidence: {nsw['confidence_score']})")

Using Utility Functions

from visonorm import detect_nsw, normalize_sentence

# Detect NSW
nsw_spans = detect_nsw("sv dh gia dinh chua cho di lam")

# Normalize sentence
normalized = normalize_sentence("sv dh gia dinh chua cho di lam")

# Normalize and detect NSW
nsw_spans, normalized = normalize_sentence("sv dh gia dinh chua cho di lam", detect_nsw=True)

4. 📊 Resource Management — Dataset Management

Datasets are stored on GitHub Releases and automatically downloaded when needed.

from visonorm import list_datasets, load_dataset, get_dataset_info

# List all available datasets
datasets = list_datasets()
print("Available datasets:")
for i, dataset in enumerate(datasets, 1):
    print(f"{i}. {dataset}")

# Get detailed information about a dataset
info = get_dataset_info("ViLexNorm")
print(f"URL: {info['url']}")
print(f"Type: {info['type']}")

# Load dataset (auto-cached)
df = load_dataset("ViLexNorm")
print(f"Dataset shape: {df.shape}")
print(df.head())

# Force re-download dataset
df = load_dataset("ViLexNorm", force_download=True)

Available datasets:

• ViLexNorm: Vietnamese Lexical Normalization Dataset
• ViHSD: Vietnamese Hate Speech Detection Dataset
• ViHOS: Vietnamese Hate and Offensive Speech Dataset
• UIT-VSMEC: Vietnamese Social Media Emotion Corpus
• ViSpamReviews: Vietnamese Spam Review Detection Dataset
• UIT-ViSFD: Vietnamese Sentiment and Emotion Detection Dataset
• UIT-ViCTSD: Vietnamese Customer Review Sentiment Dataset
• ViTHSD: Vietnamese Toxic Hate Speech Detection Dataset
• BKEE: Vietnamese Emotion Recognition Dataset
• UIT-ViQuAD: Vietnamese Question Answering Dataset

5. 🧠 Task Models — Task Processing Models

All task models are stored on HuggingFace Hub at https://huggingface.co/visolex.

SpamReviewDetection — Spam Detection

from visonorm import SpamReviewDetection

# View available models
models = SpamReviewDetection.list_models()
print("Available models:", SpamReviewDetection.list_model_names())

# Initialize with phobert-v1 model (binary classification)
spam_detector = SpamReviewDetection("phobert-v1")

# Or use other models
# spam_detector = SpamReviewDetection("phobert-v1-multiclass")  # Multiclass model

# Detect spam
text = "Sản phẩm rất tốt, chất lượng cao!"
result = spam_detector.predict(text)
print(f"Text: {text}")
print(f"Result: {result}")
# Output: Result: Non-spam

HateSpeechDetection — Hate Speech Detection

from visonorm import HateSpeechDetection

# View available models
print("Available models:", HateSpeechDetection.list_model_names())

# Initialize detector
hate_detector = HateSpeechDetection("phobert-v1")
# Or: HateSpeechDetection("phobert-v2"), HateSpeechDetection("visobert"), etc.

# Detect hate speech
text = "Văn bản cần kiểm tra hate speech"
result = hate_detector.predict(text)
print(f"Result: {result}")
# Output: Result: CLEAN

HateSpeechSpanDetection — Hate Speech Span Detection

from visonorm import HateSpeechSpanDetection

# View available models
print("Available models:", HateSpeechSpanDetection.list_model_names())

# Initialize detector
hate_span_detector = HateSpeechSpanDetection("phobert-v1")
# Or: HateSpeechSpanDetection("vihate-t5"), HateSpeechSpanDetection("visobert"), etc.

# Detect span
text = "Nói cái lồn gì mà khó nghe"
result = hate_span_detector.predict(text)
print(f"Result: {result}")
# Output: {'tokens': [...], 'text': '...'}

EmotionRecognition — Emotion Recognition

from visonorm import EmotionRecognition

# View available models
print("Available models:", EmotionRecognition.list_model_names())

# Initialize detector
emotion_detector = EmotionRecognition("phobert-v2")
# Or: EmotionRecognition("phobert-v1"), EmotionRecognition("visobert"), etc.

# Recognize emotion
text = "Tôi rất vui mừng và hạnh phúc!"
emotion = emotion_detector.predict(text)
print(f"Emotion: {emotion}")
# Output: Emotion: Enjoyment

AspectSentimentAnalysis — Aspect-based Sentiment Analysis

from visonorm import AspectSentimentAnalysis

# View available domains
print("Available domains:", AspectSentimentAnalysis.list_domains())

# View available models for a specific domain
print("Models for smartphone:", AspectSentimentAnalysis.list_model_names("smartphone"))
print("Models for restaurant:", AspectSentimentAnalysis.list_model_names("restaurant"))
print("Models for hotel:", AspectSentimentAnalysis.list_model_names("hotel"))

# Initialize with smartphone domain and phobert model
absa = AspectSentimentAnalysis("smartphone", "phobert")
# Or use other models: "phobert-v2", "bartpho", "vit5", "visobert", etc.

# Or other domains
# absa = AspectSentimentAnalysis("restaurant", "phobert-v1")
# absa = AspectSentimentAnalysis("hotel", "phobert-v1")

# Analyze sentiment
text = "Điện thoại có camera rất tốt nhưng pin nhanh hết"
aspects = absa.predict(text, threshold=0.25)
print(f"Aspects: {aspects}")
# Output: [('BATTERY', 'neutral'), ('FEATURES', 'neutral'), ('PERFORMANCE', 'positive'), ...]

6. 🎯 Advanced Usage — Advanced Usage

Combining Multiple Functions

from visonorm import BasicNormalizer, EmojiHandler, ViSoLexNormalizer

def process_text_advanced(text):
    """Process text with multiple steps"""
    print(f"Original text: {text}")
    
    # Step 1: Emoji processing
    emoji_handler = EmojiHandler()
    emojis = emoji_handler.detect_emoji(text)
    print(f"Detected emojis: {emojis}")
    
    # Step 2: Basic normalization
    normalizer = BasicNormalizer()
    normalized = normalizer.basic_normalizer(text, case_folding=True)
    print(f"Basic normalized: {normalized}")
    
    # Step 3: Lexical normalization with deep learning
    lex_normalizer = ViSoLexNormalizer()
    final_normalized = lex_normalizer.normalize_sentence(text)
    print(f"Lexical normalized: {final_normalized}")
    
    return {
        'original': text,
        'emojis': emojis,
        'basic_normalized': normalized,
        'lexical_normalized': final_normalized
    }

# Test
result = process_text_advanced("Hôm nay tôi rất😊 VUI 😊😊 và HẠNH PHÚC!")

---

🌐 Resources

HuggingFace Hub

All models and resources are published on HuggingFace Hub:

• Organization: https://huggingface.co/visolex
• Models: View full list at https://huggingface.co/visolex

Available normalization models:

• visolex/visobert-normalizer-mix100 (default)

GitHub Releases

Datasets are stored as GitHub Releases and automatically downloaded when used:

• Repository: https://github.com/AnhHoang0529/visonorm
• Releases: https://github.com/AnhHoang0529/visonorm/releases

---

📝 Citation

ViSoNorm is developed at the University of Information Technology, Vietnam National University Ho Chi Minh City (UIT, VNU-HCM). If you use ViSoNorm in your research, please cite:

@article{nguyen_weakly_2025,
	title = {A {Weakly} {Supervised} {Data} {Labeling} {Framework} for {Machine} {Lexical} {Normalization} in {Vietnamese} {Social} {Media}},
	volume = {17},
	issn = {1866-9964},
	url = {https://doi.org/10.1007/s12559-024-10356-3},
	doi = {10.1007/s12559-024-10356-3},
	number = {1},
	journal = {Cognitive Computation},
	author = {Nguyen, Dung Ha and Nguyen, Anh Thi Hoang and Van Nguyen, Kiet},
	month = jan,
	year = {2025},
	pages = {57},
}

@inproceedings{nguyen-etal-2025-visolex,
    title = "{V}i{S}o{L}ex: An Open-Source Repository for {V}ietnamese Social Media Lexical Normalization",
    author = "Nguyen, Anh Thi-Hoang  and
      Nguyen, Dung Ha  and
      Nguyen, Kiet Van",
    editor = "Rambow, Owen  and
      Wanner, Leo  and
      Apidianaki, Marianna  and
      Al-Khalifa, Hend  and
      Eugenio, Barbara Di  and
      Schockaert, Steven  and
      Mather, Brodie  and
      Dras, Mark",
    booktitle = "Proceedings of the 31st International Conference on Computational Linguistics: System Demonstrations",
    month = jan,
    year = "2025",
    address = "Abu Dhabi, UAE",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.coling-demos.18/",
    pages = "183--188",
}

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

👥 Authors

👥 Authors

• Anh Thi-Hoang Nguyen – University of Information Technology, Vietnam National University Ho Chi Minh City (UIT, VNU-HCM) – Maintainer – [e-mail]
• Ha Dung Nguyen – University of Information Technology, Vietnam National University Ho Chi Minh City (UIT, VNU-HCM) – Maintainer – [e-mail]
• Kiet Van Nguyen – University of Information Technology, Vietnam National University Ho Chi Minh City (UIT, VNU-HCM) – Maintainer – [e-mail]

---

🙏 Acknowledgments

• HuggingFace for hosting models and providing the transformers library
• The Vietnamese NLP community for datasets and feedback

---

📞 Contact & Support

• GitHub Issues: https://github.com/AnhHoang0529/visonorm/issues
• Email: [e-mail]
• HuggingFace: https://huggingface.co/visolex

---

🔗 Links

• GitHub Repository: https://github.com/AnhHoang0529/visonorm
• PyPI Package: https://pypi.org/project/visonorm/
• HuggingFace Hub: https://huggingface.co/visolex
• Documentation: https://github.com/AnhHoang0529/visonorm