IRIS — East Africa Hate Speech, Disinformation & VE Monitoring Platform (UNDP)
# MERLx IRIS — Intelligence & Risk Identification from Signals
Multi-source monitoring and classification system for hate speech, disinformation, and violent extremism narratives across **South Sudan**, **Kenya**, and **Somalia**, with dedicated **Al-Shabaab media tracking**.
Data collected via Phoenix social media gathers for UNDP Project 133.
---
## Repository Structure
```
iris/
├── README.md
├── data/
│ └── phoenix_csvs/ # Social media datasets from Phoenix gathers
│ ├── Kenya_*.csv # 6 files — Delegitimization, Diaspora, Elections, Hate Speech, Peace, VE
│ ├── Somalia_*.csv # 8 files — Delegitimization, Escalation, Hate Speech, Mixed, Peace, Unknown, VE
│ └── South_Sudan_*.csv # 6 files — Delegitimization, Diaspora, Elections, Hate Speech, Peace, VE
├── scripts/
│ ├── classify_fast.py # Main classification script (incremental, background-safe)
│ ├── run_all_models.sh # Sequential model runner with cache management
│ └── models.json # Model configurations and column mappings
├── monitoring/
│ ├── iris_protocol.md # Full monitoring protocol (Steps 1-5)
│ ├── baseline_feb26_2026.md # Threat baseline for alert decisions
│ └── findings/ # Daily scan outputs (JSON)
│ ├── README.md # Findings schema documentation
│ ├── state.json # Monitor state tracking
│ └── findings_YYYY-MM-DD.json # Daily structured findings
└── docs/
├── methodology.md # Classification methodology and data schema
└── data_dictionary.md # Complete column reference (149 columns documented)
```
## Dataset Summary
| Country | Files | Rows | Categories |
|---------|-------|------|------------|
| Kenya | 6 | 27,405 | Delegitimization, Diaspora, Elections, Hate Speech, Peace, Violent Extremism |
| Somalia | 8 | 25,070 | Delegitimization, Escalation, Hate Speech, Mixed, Mixed Monitoring, Peace, Unknown, VE |
| South Sudan | 6 | 28 …