Epidemiologia digital aplicada a arboviroses no Piauí: proposta e avaliação de uma metodologia de extração geolocalizada de menções em fontes abertas

Abstract

This work proposed, implemented as a proof of concept, and critically evaluated a Digital Epidemiology methodology. The goal was to identify and geolocate mentions of arboviruses in online public sources in the state of Piauí, emphasizing the characterization of local challenges and the analysis of the technical feasibility and limitations of the approach. It is an applied, exploratory, and descriptive study, conducted under a development research rationale with a qualitative-quantitative approach. The methodology combined three stages. The first was automated data collection (web scraping) from news outlets and official bulletins. The second was Named Entity Recognition with the spaCy library, followed by the association between disease and municipality through co-occurrence within a sentence. The third materialized the workflow in a functional prototype featuring a dashboard with an interactive map. Evaluation was both quantitative and qualitative. Quantitatively, the extracted mentions were compared against a reference set (gold standard) using precision, recall, and F1 score. Qualitatively, errors and observed limitations were analyzed. The theoretical framework drew mainly on Eysenbach, Salathé, Nadeau and Sekine, and Jurafsky and Martin. The research is justified by the limitations of official notification systems, subject to delay and underreporting, and by the gap of instruments exploring open sources in the Piauí context. The results showed high precision (94.9%) and limited recall (42.5%), with an F1 score of 58.7%. These figures reveal the feasibility of the approach as a complementary signal. At the same time, they expose relevant local challenges, such as the instability of automated collection, linguistic noise, and the heuristic nature of geolocation. It is concluded that the study contributes less as a finished solution and more as a critical characterization of the conditions, possibilities, and limits of using Digital Epidemiology over open sources for arbovirus surveillance in Piauí