Explainable epidemiological thematic features for event based disease surveillance

Edmond Odhiambo Menya, Roberto Interdonato, Dickson Owuor, Mathieu Roche

Abstract: Highlights•Training disease surveillance systems improves with biomedical language models.•Fine tuning with epidemiological data improves text based disease surveillance.•Presence of spatiotemporal features does not benefit epidemiological relevance.•Presence of keywords and biomedical features contributes the most to relevance.•Enriching thematic features improves relevance detection in disease surveillance.