README.md

metadata

language:
  - en
  - ar
  - zh
  - fr
  - ru
  - es
license: cc0-1.0
task_categories:
  - text-classification
  - token-classification
  - text-generation
  - question-answering
pretty_name: UN Security Council Complete (UNSC-Complete)
size_categories:
  - 1K<n<10K
tags:
  - legal
  - international-relations
  - voting
  - united-nations
  - diplomacy
  - geopolitics
  - multilingual
  - vetoes

UN Security Council Complete Dataset (UNSC-Complete)

🌍 The Most Comprehensive UN Security Council Dataset Available

Dataset Summary

UNSC-Complete is the first dataset to combine ALL UN Security Council activity:

• ✅ 2,722 adopted resolutions (1946-2024) with full texts and voting records
• ❌ 271 vetoed draft resolutions (1946-2025) that were blocked by P5 members
• 📊 2,993 total records showing what passes AND what fails at the UN's most powerful body

This unified dataset reveals the complete picture of Security Council decision-making, including the 90.9% passage rate and the critical 9.1% of drafts that never see the light of day due to vetoes.

🎯 Why This Dataset Matters

Previous datasets only included adopted resolutions, missing the crucial story of what gets blocked. This dataset shows:

• 271 vetoed drafts - the "dark matter" of international diplomacy
• Russia/USSR: 161 vetoes, USA: 95 vetoes, showing geopolitical fault lines
• Cold War vs Post-Cold War dynamics: 200 vetoes (1946-89) vs 71 vetoes (1990-2025)
• Complete voting records: See exactly how each country voted on every resolution

📁 Dataset Structure

Unified Schema

Every record (adopted or vetoed) contains:

{
  "unified_id": 1234,
  "res_no": 242,  // Positive for adopted, negative for vetoed
  "symbol": "S/RES/242(1967)",  // or draft number for vetoed
  "date": "1967-11-22",
  "status": "adopted",  // or "vetoed"
  "is_adopted": true,   // boolean flag

  "vote_yes": 15,
  "vote_no": 0,
  "vote_abstention": 0,
  "voting_countries": [
    {"country": "UNITED STATES", "vote": "yes"},
    {"country": "USSR", "vote": "yes"},
    ...
  ],

  "has_veto": false,
  "vetoed_by": [],  // List of P5 members who vetoed

  "english_text_best": "Full resolution text...",
  "text_length": 1977,

  "chapter7": false,  // Legal framework indicators
  "enforcement_level": "none",  // none/threat/breach/aggression

  "m49_region": "Asia",
  "cited_resolutions": ["181", "234"],
  ...
}

Key Features

Feature	Description
Unified ID	Sequential ID across all records (1-2993)
Status	"adopted" or "vetoed" - know immediately what passed
Voting Details	Complete country-by-country votes for adopted resolutions
Veto Information	Which P5 member(s) blocked each draft
Legal Framework	Chapter VI/VII/VIII, enforcement levels, human rights
Text Content	Full resolution texts (for adopted)
Geographic Scope	Countries and regions mentioned
Citation Network	Links between resolutions

📊 Dataset Statistics

Overall

• Total records: 2,993
• Success rate: 90.9% adopted, 9.1% vetoed
• Date range: 1946-2025 (79 years)
• Text length: 189 to 343,887 characters

Adopted Resolutions (2,722)

• Unanimous: 81.4%
• Chapter VII: 32.9% (enforcement actions)
• With citations: 94.5%
• Human rights: 29.1%

Vetoed Drafts (271)

• Russia/USSR: 161 vetoes (59.4%)
• United States: 95 vetoes (35.1%)
• United Kingdom: 32 vetoes (11.8%)
• China: 21 vetoes (7.7%)
• France: 18 vetoes (6.6%)
• Double/Triple vetoes: 43 drafts

Top Vetoed Topics

1. Admission of new Members (60)
2. Middle East, including Palestinian question (22)
3. Middle East (Syria) (18)
4. Occupied Arab territories (16)
5. Middle East (Lebanon) (10)

🚀 Supported Tasks

1. Veto Prediction (Binary Classification)

• Task: Predict if a draft will be vetoed
• Baseline: 9.1% positive rate (highly imbalanced)
• Challenge: Understand geopolitical red lines

2. Passage Prediction (Binary Classification)

• Task: Given draft text/metadata, predict adoption
• Baseline: 90.9% positive rate
• Value: Understand what makes resolutions passable

3. P5 Consensus Analysis

• Task: Predict P5 voting alignment
• Features: Historical patterns, topics, regional focus
• Insight: When do great powers agree?

4. Temporal Analysis

• Task: Classify era (Cold War/Post-Cold War/War on Terror/Multipolar)
• Signal: Language evolution, topics, voting patterns
• Text growth: 170 words (1946) → 3,600+ words (2011)

5. Legal Framework Detection

• Chapter VII: 32.9% of adopted (enforcement)
• Threat hierarchy: none → threat → breach → aggression
• Human rights: Growing from rare to 29.1%

6. Geographic Classification

• Multi-label: Africa (33.6%), Asia (22.2%), Europe (6.2%)
• Challenge: Multiple regions per resolution
• Vetoed drafts: Heavy Middle East focus

7. Citation Network Analysis

• Task: Predict citation links
• Data: 94.5% of adopted resolutions cite others
• Application: Understanding precedent and evolution

📥 Usage

Loading the Dataset

from datasets import load_dataset

# Load the unified dataset
dataset = load_dataset("your-username/unsc-complete")

# Separate adopted and vetoed
adopted = dataset.filter(lambda x: x['is_adopted'])
vetoed = dataset.filter(lambda x: not x['is_adopted'])

Example: Analyzing Veto Patterns

import pandas as pd

# Load the unified dataset
df = pd.read_json("unsc_unified_dataset.jsonl", lines=True)

# Analyze veto patterns over time
vetoed = df[df['status'] == 'vetoed']

# P5 veto counts
for member in ['United States', 'Russia/USSR', 'China', 'United Kingdom', 'France']:
    count = vetoed['vetoed_by'].apply(lambda x: member in x if x else False).sum()
    print(f"{member}: {count} vetoes")

# Veto rate by decade
veto_rate = df.groupby('decade').agg({
    'is_adopted': lambda x: (1 - x.mean()) * 100
}).round(1)
print(f"Veto rate by decade:\n{veto_rate}")

Example: Predicting Passage

from sklearn.ensemble import RandomForestClassifier
from sklearn.feature_extraction.text import TfidfVectorizer

# Prepare features
X_text = df['title'].fillna('') + ' ' + df['agenda_information'].fillna('')
y = df['is_adopted'].astype(int)

# Create features
vectorizer = TfidfVectorizer(max_features=1000)
X = vectorizer.fit_transform(X_text)

# Train model
model = RandomForestClassifier(class_weight='balanced')
model.fit(X, y)

# Feature importance shows what topics face vetoes
important_features = vectorizer.get_feature_names_out()[model.feature_importances_.argsort()[-20:]]
print(f"Topics associated with vetoes: {important_features}")

🗂️ Files

• unsc_unified_dataset.csv - Complete unified dataset (28.9 MB)
• unsc_unified_dataset.jsonl - JSONL format for streaming (30.8 MB)
• unsc_master_data.csv - Adopted resolutions only with full details (28.8 MB)
• unsc_vetoed_drafts.csv - Vetoed drafts only (50 KB)
• unsc_voting_details.csv - Country-by-country voting records (1.9 MB)

📚 Data Sources

1. Adopted Resolutions: CR-UNSC Academic Dataset by Fobbe, Gasbarri, and Ridi
2. Vetoed Drafts: UN DPPA Security Council Vetoes Database

⚖️ License

Released under CC0 1.0 Universal (Public Domain) in accordance with UN document policy.

📖 Citation

@dataset{unsc_complete_2024,
  title={UN Security Council Complete Dataset (UNSC-Complete)},
  author={[Your Name]},
  year={2024},
  publisher={HuggingFace},
  note={Combines adopted resolutions and vetoed drafts for complete UNSC coverage}
}

🎯 Key Insights from the Data

1. Veto Power Shapes Everything: 271 drafts never became resolutions due to P5 vetoes
2. Cold War Legacy: 73.8% of all vetoes occurred during Cold War (1946-1989)
3. Middle East Dominance: Most vetoed topic post-Cold War
4. Text Explosion: Resolutions grew from ~170 to ~3,600 words, reflecting complexity
5. Consensus Building: 81.4% of adopted resolutions are unanimous - but this hides the vetoed 9.1%

🔬 Research Applications

• International Relations: Quantify great power politics
• Conflict Studies: What conflicts get UN attention vs ignored?
• Legal NLP: Train models on diplomatic/legal language
• Temporal Analysis: 79 years of language evolution
• Network Analysis: Citation networks reveal precedent patterns
• Fairness Studies: Geographic and political biases in UN action

💡 What Makes This Dataset Unique

This is the FIRST dataset to show both sides of Security Council action:

• ✅ What passes (adopted resolutions)
• ❌ What fails (vetoed drafts)

Previous datasets only showed adopted resolutions, missing the critical story of power politics revealed by vetoes. With this complete picture, researchers can finally study:

• True P5 disagreement rates
• Topics that trigger vetoes
• Evolution of international consensus
• The "selection bias" in adopted resolutions

---

Ready to explore 79 years of international diplomacy? Download the dataset and discover what shapes our world order!