CEFR Text Analyzer

A machine learning system for classifying English text according to CEFR proficiency levels (A1-C2) using BERT-based classification.

Features

• Pure Classification Approach: No regression or contrastive learning
• Class-Weighted Loss: Handles imbalanced datasets effectively
• Mean Pooling: Better sentence representation than [CLS] token
• F1 Score Evaluation: More suitable for imbalanced classification
• REST API: Easy integration with web applications

Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Train the Model

python train.py

This will:

• Load data from dataset/dataset.csv
• Train the BERT-based classifier
• Save the best model as cefr_bert_model.pth
• Show evaluation metrics and sample predictions

3. Start the API Server

python server.py

The server will run on http://localhost:5050

API Usage

Health Check

curl http://localhost:5050/health

Predict CEFR Levels

curl -X POST "http://localhost:5050/predict" \
     -H "Content-Type: application/json" \
     -d '{"sentences": ["I like cats.", "The implementation requires careful consideration."]}'

Get CEFR Level Information

curl http://localhost:5050/levels

CEFR Levels

• A1 (Beginner): Basic everyday expressions and very simple phrases
• A2 (Elementary): Simple sentences on familiar topics and routine matters
• B1 (Intermediate): Clear communication on familiar matters and personal interests
• B2 (Upper Intermediate): Complex texts and abstract topics with good fluency
• C1 (Advanced): Flexible and effective language use for social, academic and professional purposes
• C2 (Proficient): Very high level with precise, nuanced expression and full command of language

Key Improvements

✓ Pure classification (no regression)
✓ Class-weighted loss for imbalanced data
✓ Mean pooling for better sentence representation
✓ F1 score evaluation (better for imbalanced classes)
✓ Gradient clipping for training stability
✓ Based on CEFR-SP research methodology

Files

• model.py: Main CEFR classifier implementation
• train.py: Training script
• server.py: FastAPI REST API server
• dataset/dataset.csv: Training data (required)

Requirements

• Python 3.8+
• PyTorch
• Transformers
• FastAPI
• scikit-learn
• pandas
• numpy