Do international and African outlets cover the same election as if they were two different events?
# African elections frame analysis
> Do international and African outlets cover the same election as if they were two different events?
## The question
Four recent African elections — Nigeria 2023, Kenya 2022, Senegal 2024, South Africa 2024 — were covered simultaneously by African and international press. This project asks whether the *frames* applied to those elections differ systematically by outlet origin, and along which dimensions: security, economy, democracy, identity, process, or corruption. We are **not** evaluating coverage quality or accuracy; framing analysis is descriptive, not normative. We are also **not** running a sentiment analysis — frame is about *which dimension of the story is foregrounded*, which is a different and more analytically tractable question than "is the coverage positive or negative."
The flagship deliverable here is not the headline number — it's the **pipeline + eval + analysis modularity**: an LLM-classification system whose outputs have been validated against a hand-labeled gold set.
## Data
| Source | Granularity | Time coverage | Access |
|--------|-------------|---------------|--------|
| GDELT 2.0 Global Knowledge Graph | Article-level | ±30 days around each election | Public (raw archives) |
| Election dates & outlet provenance | Per outlet | Hand-curated for this project | Curated in `data/external/outlets.csv` |
| Hand-labeled evaluation set | Article-level | 200–300 articles, stratified | **Labeled by Muhanad** — gold set, not LLM-generated |
Access date: 2026-05 — GDELT 2.0 GKG archives pulled for the four ±30-day windows; the exact file IDs are committed in `data/raw/manifest.json`.
## Method
A four-stage pipeline. **(1) Ingestion** pulls GDELT GKG records for the ±30-day window around each of the four elections. **(2) Cleaning** filters to election-relevant English-language articles, deduplicates, and attributes each outlet to "African" or "International" using a hand-curated provenance list. **(3) Classif …