Uzbek Telegram-Style Messages for Spam Detection

Featuring 2000 Uzbek text messages labeled as 'spam' or 'normal'. It is designed for training spam detection models, with a split of 1800 training and 200 test samples.

Use Cases

• Train a text classifier to predict the 'spam' or 'normal' label for Uzbek messages.
• Analyze linguistic patterns in 'spam' messages to identify common features of aggressive advertising.
• Benchmark spam detection models on a dataset containing messages in both Latin and Cyrillic scripts.

Strengths

• Contains 2000 labeled text samples for model training and evaluation.
• Provides a predefined split of 1800 training and 200 test samples.
• Covers messages in the Uzbek language using both Latin and Cyrillic scripts.

Limitations

• The dataset is synthetic, which may not fully represent real-world message distribution or noise.
• With only 2000 total samples, the dataset is relatively small for training complex deep learning models.
• The description notes it targets 'Telegram-style' messages, which may limit generalizability to other platforms.

Provenance

Source

huggingface

Collection Method

Synthetic generation for spam detection training.

Freshness

Last updated on 2025-12-18.

Geography

Uzbekistan