Summary
Social media posts during disasters hold important information for emergency responders, but most existing data only labels entire posts rather than the specific details inside them. The authors created HUMAID-NER, a new dataset that tags around 175,000 important words and phrases, such as locations or damage reports, within 60,000 disaster-related tweets. They also built a computer model that can recognize these details and categorize the tweets by event type at the same time, using a technique to balance the two tasks for better accuracy. This work helps to quickly turn many short social media posts into structured facts that aid humanitarian efforts.
What this means in practice
- •For emergency response teams: Automatically extract detailed location and damage reports from disaster tweets to speed up humanitarian decision making.
- •For social media monitoring platforms: Incorporate joint entity and event recognition features tailored for crises to improve disaster-related information tracking.
Authors
Aijaz Ali, Nazish Basir, Sarfaraz Nawaz, Danish Nazir Arain, Haris Ali
Abstract
Rapid extraction of structured information from social media is important for humanitarian response, yet existing disaster tweet resources mainly provide document-level category labels without span-level entity annotations. We introduce HUMAID-NER, the first named entity recognition dataset built on the HumAID benchmark, containing 60,000 English disaster tweets annotated in BIO format across ten operationally motivated entity types and yielding approximately 175,000 labelled entity spans. Annotations are generated through a reproducible three-stage hybrid pipeline combining a spaCy transformer model, disaster-domain EntityRuler patterns, and structured regular expressions with priority-based overlap resolution. We also propose a joint multitask learning framework that performs disaster-specific named entity recognition and humanitarian event classification using a shared RoBERTa-large encoder. To reduce task conflict during joint training, the model uses homoscedastic uncertainty weighting with learnable task parameters and a two-stage training schedule that freezes the lower 18 of 24 encoder layers in the second stage. On the HUMAID-NER validation set, the proposed system achieves NER span micro-F1 of 0.841 and classification macro-F1 of 0.761 simultaneously. A real-time web dashboard demonstrates end-to-end deployment. The dataset, models, and pipeline code are released to support reproducibility and future crisis informatics research.