Fake News Radar

A polished Streamlit app for fake news classification with a single-page tabbed UX, URL content extraction, and profile-based model training on the WELFake dataset.

Highlights

• Single-page tab navigation (Home, Text Analyzer, URL Analyzer, Model Insights) in app.py
• Text classification with confidence scoring
• URL analysis using both article title and main body content
• WELFake training profiles:
  • quick: stratified subset for faster training
  • full: full dataset training
• Automatic dataset download from Google Drive when training starts (no large CSV committed to GitHub)
• Profile-specific saved artifacts with active profile switching
• Model metrics dashboard (Accuracy, Precision, Recall, F1, ROC-AUC, confusion matrix)

Tech Stack

• Python
• Streamlit
• scikit-learn
• pandas / numpy
• matplotlib
• BeautifulSoup + requests

Repository Layout

fake-news-streamlit-app/
|-- app.py
|-- artifacts/
|   |-- datasets/
|   |   `-- WELFake_Dataset.csv   # downloaded at runtime on first training run
|   |-- fake_news_model_quick.joblib
|   |-- fake_news_model_full.joblib
|   |-- tfidf_vectorizer_quick.joblib
|   |-- tfidf_vectorizer_full.joblib
|   |-- training_metrics_quick.json
|   |-- training_metrics_full.json
|   |-- training_metrics_active.json
|   `-- active_profile.txt
|-- src/
|   |-- models/
|   |   |-- predict.py
|   |   `-- train.py
|   |-- ui/
|   |   |-- components.py
|   |   `-- theme.py
|   `-- utils/
|       `-- web_scraper.py
|-- requirements.txt
`-- README.md

Setup

1. Clone the repo

git clone <your-repo-url>
cd fake-news-streamlit-app

2. Install dependencies

pip install -r requirements.txt

3. Run the app

streamlit run app.py

How Training Works

Use the Model Insights tab and click Train on WELFake with either quick or full selected.

If the dataset is not already cached locally, the app downloads it automatically from Google Drive:

• https://drive.google.com/file/d/13lcNYSvVfJhC5xl-84k5AcHvNVnKiI1T/view?usp=drive_link

The downloaded file is cached at:

• artifacts/datasets/WELFake_Dataset.csv

You can override the dataset URL in deployment environments using:

• WELFAKE_DATASET_URL

• Quick profile writes:

  • artifacts/fake_news_model_quick.joblib
  • artifacts/tfidf_vectorizer_quick.joblib
  • artifacts/training_metrics_quick.json

• Full profile writes:

  • artifacts/fake_news_model_full.joblib
  • artifacts/tfidf_vectorizer_full.joblib
  • artifacts/training_metrics_full.json

The active profile is tracked in:

• artifacts/active_profile.txt
• artifacts/training_metrics_active.json

Inference always uses the active profile artifacts.

Label Semantics

The WELFake label mapping used in training/inference is:

• 0 = REAL
• 1 = FAKE

Notes for GitHub Push

• This repo is configured to ignore the artifacts/ folder by default so trained model files are not uploaded unintentionally.
• WELFake_Dataset.csv does not need to be committed. It is downloaded at runtime from Google Drive during training.
• Ensure the Google Drive file is shared as Anyone with the link (Viewer) for Streamlit Cloud deployments.

License

MIT