WikiBias

A tool for analyzing Wikipedia articles to detect various forms of bias and verify factual accuracy through automated source checking.

⚠️ Note: This is a prototype. Currently only supports analyzing 1 paragraph at a time.

What it does

WikiBias analyzes Wikipedia pages by:

• Detecting 15+ types of textual bias (loaded language, framing bias, missing context, etc.)
• Verifying claims against their cited sources by scraping and analyzing the actual content
• Evaluating source quality, diversity, and clustering
• Generating bias and factuality scores with detailed explanations

Setup

1. Clone the repository and install dependencies:

pip install -r requirements.txt

2. Configure your environment:

cp .env.sample .env

Edit .env to set up your LLM provider:

• For local LLM: Set LOCAL=True and configure API_BASE (e.g., for LM Studio)
• For OpenAI: Set LOCAL=False and add your OPENAI_API_KEY

Usage

Analyze a Wikipedia article:

python main.py "Stalin" --max-paragraphs 1

Options:

• --max-paragraphs N: Limit analysis to first N paragraphs (currently max 1)

Output

The tool generates a comprehensive JSON report including:

• Paragraph-level bias findings and source analyses
• Overall bias score (0-10)
• Overall factuality score (0-10)
• Political alignment detection
• Detailed explanations for each finding