This repository contains resources for the mini-research project, "A Model for Addressing AI Algorithm Bias through Indigenous South African Knowledge Systems". The study integrates Indigenous Knowledge Systems into AI frameworks to reduce bias and enhance cultural sensitivity in AI-generated conten
**Overview**
This Streamlit application performs pairwise agreement analysis to measure consistency between pairs of AI models (ChatGPT-Gemini, ChatGPT-Mistral AI, Gemini-Mistral AI) on plant classification tasks. The application provides detailed model-to-model comparison and reveals patterns of systematic agreement or disagreement.
Features
Pairwise Analysis: Calculates agreement rates between all model pairs
Multiple Visualization Types
1. Pairwise agreement heatmap
2. Model performance comparison chart
3. Plant-level agreement patterns
**Detailed comparison matrix**
- Flexible Data Input
- Pre-loaded research data
- User-provided classification data
**Advanced Analytics**
1. Best/worst performing model pairs
2. Overall agreement metrics
3. Statistical interpretation
***Dependencies**
# Core dependencies
pip install streamlit
pip install pandas
pip install numpy
# Visualization dependencies
pip install matplotlib
pip install seaborn
# Statistical analysis
pip install scipy
# Enhanced colormaps
pip install matplotlib-colors
**How to run the app local**
1. step 1- install dependencies
pip install -r requirements_pairwise.txt
2. step 2- Run the application
streamlit run pairwise_agreement_app.py
3. Access the application
Open browser to
localhost
Application loads with study data analysis by default
**Usage instructions**
1. Study Data Analysis
Default view displays research data with 20 indigenous plants
Click "Run Pairwise Agreement Analysis on Study Data"
View comprehensive results
- Pairwise agreement heatmap (color-coded matrix)
- Detailed comparison table with counts and percentages
- Model performance bar chart
- Plant-level agreement patterns
2. Custom Data Analysis
- Select "Insert Your Own Data" in navigation
- Input requirements
- Minimum 3 plants (fewer required than Fleiss Kappa)
- Unique plant names only
- Complete classifications for all three models
- Run analysis to get personalized pairwise results
Output Inte …