Event Detection from Social Media for Epidemic Preparedness

Anonymous

Abstract: Social media is an easy-to-access platform providing timely updates about societal trends and events. Discussions regarding epidemic-related events such as infections, symptoms, and locally deployed measures can be crucial for policy making during epidemic outbreaks. In this work, we exploit Event Detection (ED) for extracting and capturing relevant events from social media posts to provide better preparedness for any upcoming epidemic. To facilitate this task, we curate an epidemic event ontology comprising seven generic event types such as infect, symptom, prevent, etc. Using our event ontology and human expert annotation, we construct our epidemic preparedness Twitter dataset SPEED comprising 1,975 tweets and 2,217 event mentions for the COVID-19 pandemic. Experiments reveal that existing ED models and datasets cannot transfer well for our task, highlighting the challenging nature of our dataset. Finally, we provide empirical evidence highlighting the utility and generalizability of our dataset by showing that ED models trained on our COVID-only dataset SPEED, can effectively identify epidemic events and offer timely warnings for three unseen epidemics of Monkeypox, Zika, and Dengue. This generalizability of SPEED lays the foundations for better preparedness against emerging epidemics.

Paper Type: long

Research Area: Information Extraction

Contribution Types: Data resources

Languages Studied: English

Consent To Share Submission Details: On behalf of all authors, we agree to the terms above to share our submission details.