Fake News Detection with BERT

This project implements a fake news detection system using BERT and PyTorch. The model classifies news articles as either FAKE or REAL.

Features

• Utilizes the bert-base-uncased model from HuggingFace Transformers.
• Custom NewsDataset class for efficient data loading and tokenization.
• 5-Fold Cross Validation for robust evaluation.
• Training and validation loops with accuracy reporting.

Project Structure

• ui_bert.py: Main script for data loading, model definition, training, and evaluation.
• full_dataset.csv: CSV file containing news articles and labels.

Usage

1. Install dependencies:

   pip install pandas torch scikit-learn transformers

2. Prepare your dataset:

   • Ensure full_dataset.csv is in the project directory.
   • The CSV should have columns: text (news content) and label (FAKE or REAL).

3. Run the script:

   python ui_bert.py

Model Overview

• Tokenizer: BERT tokenizer (bert-base-uncased)
• Model: BERT with a frozen encoder and a linear classification head
• Loss: CrossEntropyLoss
• Optimizer: Adam

Cross Validation

The script uses 5-fold cross validation to split the dataset and reports accuracy for each fold.

Custom Classes

• NewsDataset: Handles tokenization and formatting for BERT input.
• FakeNewsClassifier: Wraps BERT and adds a linear layer for classification.

Notes

• BERT parameters are frozen during training for efficiency.
• Adjust n_epochs and batch_size in ui_bert.py as needed.

License

This project is for educational purposes.