This study introduces a novel two-stage supervised machine-learning algorithm designed to improve geolocation accuracy in event data.
Data & Methods
- Extracts contextual information from texts including N-gram patterns for location words, their mention frequency, and surrounding sentence context.
- Uses training datasets (customized from news articles globally) to estimate model parameters.
- Employs the trained model on test data to predict if a location word correctly represents an event's actual place.
Key Findings
- The algorithm successfully identifies inaccuracies in location mentions by analyzing surrounding text.
- It demonstrates superior performance compared to existing geocoders, even when processing unseen news articles.
Why This Matters for Political Science
- Accurate event geolocation is crucial for tracking political phenomena and trends across countries.
- This approach provides a reliable method for enhancing the precision of automated text analysis in political research.
