Political Leaning Detection in News Headlines

This project aims to analyze political bias in mainstream media by classifying news articles from various outlets as right-leaning, centrist, or left-leaning. The goal is to assess potential biases within these news sources and provide insights into their political orientations. The implementation leverages transformer for sequence classification, particularly fine-tuning BERT with Low-Rank Adaptation (LoRA). Post-training quantization (PTQ) was used to optimize the model’s performance.

Possible Use Cases

Helping people understand news bias

• Classifies news articles to show if they lean left, right, or center, and 
• Aims to help readers make informed choices by identifying bias in news sources

Supporting NGOs and journalists with media monitoring

• Monitors bias trends on specific issues, helping ensure balanced reporting
• Useful for detecting potential misinformation or slanted perspectives

Assisting academic research

• Useful for researchers studying media bias, journalism, or political influence in the news

The Dataset

Released on 15 July 2020, the POLUSA dataset contains 0.9 million political news articles, carefully balanced across different periods and news outlet popularity. It provides a valuable resource for analyzing political trends and biases in media. The dataset is available for download on Zenodo.org.

Link: https://zenodo.org/records/3946057/files/polusa_balanced.zip?download=1

Setup

Follow these steps to set up the project:

Clone the repository

git clone https://github.com/iampujan/political_leaning_news_detection_backend.git
cd your-repository

Install UV Python package and project manager

For macOS and Linux

curl -LsSf https://astral.sh/uv/install.sh | sh

For Windows

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Install dependencies

uv sync

Usage

Run the notebook sequentially in a Jupyter Notebook environment or a similar setup:

• Step 1: Download data using gdown or any alternative method.
• Step 2: Preprocess text data with tokenization, stopword removal, and bigram extraction.
• Step 3: Fine-tune a transformers model using the provided pipeline.
• Step 4: Evaluate model performance and generate classification reports.

Reproducibility

To reproduce the results:

1. Ensure the dependencies are installed as described in the Setup section.
2. Follow the cell execution sequence in the notebook:
   • Data download and exploration.
   • Data preprocessing.
   • Model training and evaluation.
3. Save results and logs using mlflow.

Running the Application

Backend (FastAPI)

uvicorn app.main:app --host 0.0.0.0 --port 8080

The server should start http://localhost:8080

Frontend (React)

npm start

The server should start on http://localhost:3000

Contact

For any questions or clarifications, please contact Raza Mehar at [raza.mehar@gmail.com], Pujan Thapa at [iampujan@outlook.com] or Syed Najam Mehdi at [najam.electrical.ned@gmail.com].