This script calculates BERTScore for Arabic sentences using a pre-trained Arabic BERT model. It provides an interactive interface for users to input sentence pairs directly or process multiple pairs from a CSV file.

• Calculate BERTScore for Arabic sentence pairs
• Interactive command-line interface
• Support for direct input of sentence pairs
• Support for batch processing using CSV files
• Detailed output including individual and average scores

• Python 3.6 or higher
• PyTorch
• Transformers library
• BERTScore library

1. Clone this repository or download the arabic_bertscore_calculator.py script.

2. Install the required Python libraries:

   pip install torch transformers bert-score
   

Run the script using Python:

python arabic_bertscore_calculator.py

Follow the on-screen prompts to choose your input method:

1. Choose option 1 when prompted.
2. Enter the reference sentence when asked.
3. Enter the candidate sentence when asked.

The script will calculate and display the BERTScore metrics for this sentence pair.

1. Choose option 2 when prompted.
2. Enter the path to your CSV file when asked.

The script will process all sentence pairs in the CSV file and display the results.

• The CSV file should have two columns: the first for reference sentences and the second for candidate sentences.
• Each row represents one sentence pair.
• Use UTF-8 encoding to ensure proper handling of Arabic text.

Example CSV content:

مرحبا كيف حالك؟,كيف حالك؟ مرحبا
الجو جميل اليوم,اليوم الطقس رائع

The script will display:

• BERTScore metrics (Precision, Recall, F1) for each sentence pair
• The reference and candidate sentences for each pair
• Average scores across all processed pairs

• This script uses the "asafaya/bert-base-arabic" model for BERTScore calculations.
• The scores are not rescaled with a baseline, which may affect interpretability across different models or languages.

• If you encounter a "module not found" error, ensure all required libraries are installed.
• For CSV file issues, check that your file is properly formatted and uses UTF-8 encoding.
• If you face CUDA-related errors, ensure your PyTorch installation is compatible with your CUDA version, or the script will default to CPU.

Feel free to fork this repository and submit pull requests for any enhancements.

This project is open-source and available under the MIT License.