vsem-azamat/rubert-tiny-spam-classifier

RuBERT Tiny Spam Classifier 🤖

A lightweight Russian language spam classifier based on RuBERT Tiny model. The model can detect spam in text messages with high accuracy while maintaining minimal resource requirements.

📋 Requirements

pip install --upgrade pip
pip install -r requirements.txt

🗂️ Project Structure

.
├── cleaned_dataset.csv     # Cleaned dataset
├── messages.csv            # Source dataset
├── model/                  # Trained model
├── scripts/
│   ├── inference.py        # Inference script
│   ├── preprocess.ipynb    # Data preprocessing notebook
│   └── train.py            # Training script
└── requirements.txt        # Project dependencies

Datasets will not be published at this time.

🎯 Usage

Train

python3 scripts/train.py

Inference

# python3 scripts/inference.py
from scripts.inference import SpamClassifier

classifier = SpamClassifier(model_path="./model")

text = "Привет, у меня есть подработка, оплата 100 сек руб.
is_spam = classifier.classify(text)
print(f"Is spam: {is_spam}")

📈 Metrics

• Accuracy: 0.99
• Precision: 0.89
• Recall: 0.96
• F1 Score: 0.92

🔍 Model

The classifier is based on RuBERT Tiny - a lightweight version of RuBERT, optimized for running on low-resource machines. The model is fine-tuned on a dataset of Russian messages for spam classification.

📝 License

MIT License