HUMAID-NER Dataset for Disaster Event and Entity Extraction
September 16, 2026
The HUMAID-NER dataset provides 60,000 English disaster tweets with 175,000 annotated entity spans across ten categories. It uses a joint multitask learning framework with a RoBERTa-large encoder to perform simultaneous named entity recognition and event classification.
HOW THIS AFFECTS YOU
●
builderYou can integrate this multitask framework to improve structured information extraction from social media feeds.
●
researcherYou can use this new span-level annotation dataset to train more precise disaster-response models.