Machine learning–based estimation of DPRK recommendation acceptance patterns in the Universal Periodic Review (UPR) process.
This project uses historical Universal Periodic Review (UPR) recommendation data and machine learning models to estimate the probability that the Democratic People’s Republic of Korea (DPRK) will accept or reject a recommendation during the UPR process.
The system combines political, thematic, and discursive variables to generate probabilistic predictions based on historical behavioral patterns observed in previous UPR cycles.
The tool is designed for:
- academic research,
- policy analysis,
- human rights monitoring,
- and exploratory scenario simulation.
⚠️ Predictions generated by this application are probabilistic and should be interpreted as analytical scenarios rather than deterministic forecasts.
- 🌐 Bilingual interface (English / Spanish)
- 🤖 Machine learning–based prediction engine
- 🧭 Political sensitivity classification
- 🏛️ Thematic area classification
- 📊 Alignment score integration using UNGA voting data
- 📝 Recommendation text analysis
- ⚖️ Scenario simulation modes
- 📈 Probabilistic prediction outputs
- 🎛️ Interactive Streamlit dashboard
The model estimates DPRK response behavior using structured variables derived from UPR recommendations and political indicators.
| Variable | Description |
|---|---|
| Political Sensitivity | Degree of regime sensitivity associated with the recommendation |
| Thematic Area | Main substantive issue addressed |
| Alignment Score | Ideological proximity between the recommending State and the DPRK |
| Recommendation Length | Discursive complexity and density |
| Recommendation Text | Semantic content of the recommendation |
| Recommending State | Country issuing the recommendation |
The project operationalizes political sensitivity using a three-level analytical framework.
| Level | Description |
|---|---|
| Low Sensitivity | Social and humanitarian issues generally tolerated by the regime |
| Medium Sensitivity | Institutional or normative issues implying moderate political pressure |
| High Sensitivity | Issues directly linked to political control, repression, accountability, or civil liberties |
- food
- health
- education
- disability rights
- humanitarian cooperation
- gender equality
- legal reform
- treaty ratification
- institutional strengthening
- torture
- prison camps
- executions
- freedom of expression
- political participation
- ICC / COI references
The alignment score captures ideological proximity using United Nations General Assembly voting patterns, following the literature that treats voting behavior as a proxy for state preferences (Voeten, 2013).
In this project, the score is operationalized as the average absolute voting distance between each State and the DPRK:
- 0 = no ideological proximity
- 1 = complete alignment
The application includes different analytical behavior modes:
| Scenario | Description |
|---|---|
| 50/50 Coin Flip Dynamic | Assumes balanced uncertainty |
| 80/20 Pareto Logic | Assumes asymmetric behavioral concentration |
These scenarios are designed to explore different predictive distributions and model assumptions.
- Python
- Streamlit
- scikit-learn
- pandas
- numpy
- plotly
Clone the repository:
git clone https://github.com/marialasa/dprk-upr-prediction-tool.git
cd dprk-upr-prediction-toolInstall dependencies:
pip install -r requirements.txtRun the application:
streamlit run app.py├── app.py
├── data/
├── models/
├── assets/
├── notebooks/
├── requirements.txt
└── README.md
This tool was developed for:
- academic exploration,
- human rights research,
- policy analysis,
- and computational social science applications.
It is not intended to produce definitive predictions about DPRK behavior.
This project does not predict future DPRK behavior with certainty. Outputs should be interpreted as probabilistic analytical scenarios derived from historical patterns observed in prior UPR cycles.
The classifications and analytical categories used in this project are simplified operationalizations created for modeling purposes and do not represent official United Nations taxonomies.
- SHAP explainability integration
- Transformer-based NLP models
- Temporal modeling across UPR cycles
- Multilingual recommendation embeddings
- Recommendation clustering analysis
- Confidence interval estimation
If you use this project in academic work, please cite:
@software{dprk_upr_prediction_tool,
title={DPRK UPR Response Prediction Tool},
author={LASA, María de los Ángeles},
year={2026},
url={https://uprnorthkorea.streamlit.app/}
}MIT License