30,000 messages categorized into 36 disaster response classes, sourced from major global events like the 2010 Haiti earthquake and 2012 Super-storm Sandy. The collection includes direct communications and news articles spanning multiple years and hundreds of distinct disaster scenarios.
Use Cases
- Train multi-label text classifiers to map raw messages to the 36 encoded disaster response categories
- Evaluate model performance on domain-specific terminology found in messages from the 2010 Haiti earthquake and 2010 Pakistan floods
- Build a prioritization engine for emergency services using the 36 response-related labels to identify urgent needs in news articles and direct communications
Strengths
- 30,000 messages covering specific events like the 2010 Haiti earthquake and 2010 Pakistan floods
- 36 distinct classification categories for disaster response mapping
- Multi-source data including direct messages and news articles from hundreds of different disasters