sukhrobnurali/uzbek_spam_dataset

Uzbek Spam Detection Dataset

A synthetic dataset for training spam detection models for Uzbek language, specifically targeting Telegram-style messages.

Dataset Description

This dataset contains 2000 Uzbek text messages labeled as either "spam" or "normal".

Dataset Summary

• Total samples: 2000
• Train split: 1800
• Test split: 200
• Labels: spam, normal
• Language: Uzbek (Latin and Cyrillic scripts)

Spam Categories Covered

• Aggressive advertising and promotions
• Get-rich-quick schemes
• Fake job offers
• Fake giveaways and prizes
• Channel/group promotion spam
• MLM/pyramid scheme promotions
• Clickbait messages
• Loan/credit advertisements
• Too-good-to-be-true offers

Normal Message Categories

• Casual greetings
• Work/business messages
• Family conversations
• Making plans
• Asking questions
• Sharing news/opinions
• Friendly everyday conversations

Usage

from datasets import load_dataset

dataset = load_dataset("sukhrobnurali/uzbek_spam_dataset")

# Access train split
train = dataset["train"]
print(train[0])
# {'text': 'Assalomu alaykum, qalaysiz?', 'label': 'normal'}

# Access test split
test = dataset["test"]

Data Format

Each sample contains:

• text (string): The message content
• label (string): Either "spam" or "normal"

Generation Method

This dataset was synthetically generated using Google Gemini API with carefully crafted prompts to produce realistic Uzbek Telegram messages. The generation focused on common spam patterns observed in Uzbek-speaking Telegram communities.

Intended Use

• Training spam detection models for Uzbek language
• Research on low-resource language NLP
• Building content moderation systems for Uzbek platforms
• Educational purposes for understanding spam patterns

Limitations

• Synthetically generated (may not cover all real-world spam patterns)
• Based on common spam patterns as of 2025
• May need augmentation with real-world data for production use
• Balanced dataset (real-world spam distribution may differ)

License

MIT License - Free for research and commercial use.

Author

Created by Sukhrob Nurali

Citation

@dataset{uzbek_spam_dataset_2025,
  title={Uzbek Spam Detection Dataset},
  author={Sukhrob Nurali},
  year={2025},
  publisher={HuggingFace},
  url={https://huggingface.co/datasets/sukhrobnurali/uzbek_spam_dataset}
}

## Contributing

If you have real-world Uzbek spam examples to contribute, feel free to open a discussion or PR to help improve this dataset.