SP25-690-Veeram-Reddy

Fake News Detection Using Transformer-Based Deep Learning Models

Overview

This project compares traditional machine learning, basic neural networks, and transformer-based deep learning models for binary fake-news detection.

Dataset

Dataset: davanstrien/WELFake from Hugging Face.
Task: classify news articles as fake or real.

Models

1. TF-IDF + Logistic Regression
2. Embedding MLP
3. BiLSTM
4. DistilBERT fine-tuning

Evaluation

Primary metric: macro F1-score.
Additional metrics: accuracy, precision, recall, ROC-AUC, confusion matrix.

Controlled comparison / ablation

The notebook compares body-only input against title+body input using the same TF-IDF Logistic Regression setup.

How to run

1. Open the notebook in Google Colab.
2. Select GPU runtime.
3. Run all cells from top to bottom.
4. Adjust MAX_ROWS, EPOCHS_NN, and EPOCHS_TRANSFORMER for final experiments.

Reproducibility

Random seed: 42.

Repository structure

.
├── COMP_690_AH2_Final_Draft.ipynb
├── README.md
├── requirements.txt

Responsible use

This model should be used only as a decision-support or triage tool. It should not be treated as an automatic truth authority.