BERT-Based AI Content Classification

This project leverages a BERT-based deep learning model to classify text articles as either AI-generated or human-written. Using PyTorch and the Hugging Face transformers library, the project implements fine-tuning of a pre-trained BERT model for binary classification.

---

Overview

The primary goal of this project is to classify text data into two categories:

• AI-generated
• Human-written

The workflow includes:

1. Tokenizing text data using a BERT tokenizer.
2. Defining a PyTorch dataset and data loader for text and labels.
3. Building and training a custom BERT-based classifier.
4. Evaluating the model using stratified cross-validation.
5. Saving the trained model for deployment or further analysis.

---

Features

• Pre-trained BERT Model: Fine-tunes bert-base-uncased for text classification.
• Custom Dataset Class: Implements a PyTorch-compatible dataset class for efficient data handling.
• Cross-Validation: Uses Stratified K-Fold cross-validation to ensure robust evaluation.
• Evaluation Metrics: Calculates accuracy, F1 score, precision, and recall.

---

Requirements

• Python 3.7+
• Libraries:
  • torch
  • transformers
  • pandas
  • numpy
  • sklearn

Future Improvements

• Experiment with advanced pre-trained models like RoBERTa or DeBERTa.
• Handle class imbalance with techniques like oversampling or weighted loss functions.
• Extend the model to support multiclass classification for other types of text.