Overview

This project presents the first Arabic AI detector developed after the emergence of ChatGPT. It's designed to differentiate between Arabic synthetic ChatGPT texts and human-written texts, addressing critical gaps in current AI detection systems in dealing with this language. This code is customizable to adapt to any language. The only thing that needed to be adjusted is the Dediacritization Layer, which is designed specifically for the Arabic language.

Our Design

• Customizable: Easily configurable via the config.json file containing all necessary hyperparameters.
• Inference Module: Utilize ModelEvaluator.py if you want to investigate the performance for inference, leveraging the best model weights and Testing dataset of your choice.

Installation of Dependencies

Usage Instructions

pip install -r requirements.txt

Configuration

1. Modify Configurations: Edit the config.json file to set the model hyperparameters as per your requirements.

Model Weights

1. Download Models: Access the Links To Experiments.txt in the Results folder to download the model weights.
2. Place Weights: Ensure the downloaded weights are placed in the check_point folder for the inference module to function correctly.

Dataset Preparation

1. File Naming: Edit the prefix of the dataset name. Our datasets are prefixed with labels like customTraining, largeTraining, etc. Rename these to Training.csv, Validation.csv, and Testing.csv for operational compatibility, and they have prefixed for readability purposes and organization within the dataset.