🔬 Aspect Extraction for Scientific Texts

Multilingual Aspect Extraction with Transfer Learning: Russian → Kazakh

This project implements XLM-RoBERTa + CRF model for aspect extraction in scientific texts, with focus on zero-shot transfer learning from Russian to Kazakh.

🎯 Research Goal

Transfer learning approach to extract scientific aspects (AIM, METHOD, MATERIAL, TASK, TOOL, RESULT, USAGE) from Kazakh scientific texts using Russian training data.

📊 Data for Experiments

Data used for experiments can be in the original repository of SciMDIX Dataset: https://github.com/tvbat/sci-text-miner-scimdix/tree/main

🚀 Quick Start

For GPU Server (Recommended)

# Clone and setup
git clone <repository>
cd scimdix_aspect_extraction
chmod +x scripts/*.sh

# Quick test (1 epoch)
scripts/run_single_experiment.sh test

# Main zero-shot experiment
scripts/run_single_experiment.sh zero_shot_ru_to_kz

# Run all experiments
scripts/run_all_experiments.sh

For Local Development

# Setup environment
python -m venv venv
source venv/bin/activate  # or `venv\Scripts\activate` on Windows
pip install -r requirements.txt

# Test locally
PYTHONPATH=. python scripts/test_cpu_training.py

# Run experiment natively
scripts/run_native.sh test

📊 Experiments

• baseline_ru - Russian baseline (ru→ru)
• baseline_kz - Kazakh baseline (kz→kz)
• zero_shot_ru_to_kz - Main experiment: Russian→Kazakh transfer
• lodo_* - Leave-One-Domain-Out validation (IT, linguistics, medical, psychology)

🏗️ Architecture

• Model: XLM-RoBERTa-base + Linear + CRF
• Training: Dual learning rates (encoder: 2e-5, head+CRF: 1e-4)
• Evaluation: Span-level precision, recall, F1 with exact matching
• Multi-seed: [13, 21, 42] for statistical significance

📁 Project Structure

├── src/                    # Core source code
│   ├── data/              # Data preparation pipeline
│   └── model/             # Model, trainer, evaluator
├── scripts/               # Execution scripts
│   ├── run_single_experiment.sh
│   ├── run_all_experiments.sh
│   ├── run_native.sh
│   └── test_cpu_training.py
├── docker/                # Docker configurations
│   ├── Dockerfile
│   ├── Dockerfile.simple
│   └── docker-compose.yml
├── docs/                  # Documentation
├── datasets/              # Training and test data
│   ├── raw/              # Original CSV files
│   └── prepared/         # Processed CoNLL files
└── results/              # Experiment outputs
    ├── experiments/      # Metrics and logs
    ├── models/          # Saved model weights
    └── logs/            # Training logs

🔧 Configuration

Key parameters in src/model/config.py:

• Batch size: 32 (GPU) / 1-2 (CPU)
• Max sequence length: 384 tokens
• Epochs: 20 for full training
• Early stopping: 3 epochs patience
• GPU: Configured for GPU=1 by default

📈 Expected Results

The system evaluates on 7 aspect classes with span-level metrics:

• Micro-averaged F1: Overall performance
• Macro-averaged F1: Performance across all classes
• Per-class metrics: Precision, recall, F1 for each aspect
• Confusion matrix: Error analysis

📚 Documentation

• Docker GPU Guide - Detailed deployment instructions
• Data Preparation - Data processing pipeline
• Model Architecture - Technical details

🛠️ Troubleshooting

Docker Issues

# Try simple Dockerfile
docker build -f docker/Dockerfile.simple -t aspect-extraction .

# Or run natively
scripts/run_native.sh test

GPU Configuration

# Check GPU availability
nvidia-smi

# Set specific GPU
export CUDA_VISIBLE_DEVICES=1

📄 Citation

If you use this code in your research, please cite:

@misc{scimdix_aspect_extraction,
  title={Multilingual Aspect Extraction for Scientific Texts: Russian-Kazakh Transfer Learning},
  author={Your Name},
  year={2025},
  note={Research implementation for scientific text analysis}
}

🤝 Contributing

1. Fork the repository
2. Create feature branch: git checkout -b feature-name
3. Commit changes: git commit -am 'Add feature'
4. Push to branch: git push origin feature-name
5. Submit pull request

📧 Contact

For questions about the research or implementation, please open an issue or contact us.

---

Status: ✅ Ready for GPU training | 🧪 Tested on RTX 3090 | 🐳 Docker deployed