Desenvolvimento de um modelo epidemiológico aprimorado com dados de redes sociais e processamento de linguagem natural para a predição de casos de COVID-19

Abstract

Introduction: Este trabalho aborda os desafios enfrentados pelos sistemas de saúde no Brasil durante a pandemia de COVID-19, com foco no papel das mídias sociais no apoio à saúde pública durante a crise pandêmica. Ressalta-se que, com a necessidade de isolamento físico e distanciamento social para reduzir a propagação do novo coronavírus durante a pandemia de COVID-19, a sociedade tem se voltado cada vez mais para os ambientes virtuais, levando a um aumento significativo do uso das redes sociais. Consequentemente, a análise de informações das mídias sociais ganhou considerável relevância, impulsionando o uso e desenvolvimento de diversas técnicas de inteligência artificial para criar modelos preditivos. Assim, a utilização desses modelos tornou-se uma ferramenta relevante para potencializar ações no combate a surtos, epidemias, pandemias e outras situações críticas de saúde. Objetivo: Esta pesquisa teve como objetivo desenvolver um modelo epidemiológico estatístico bayesiano que incorporou dados de mídia social e processamento de linguagem natural para prever casos de COVID-19. Objective: The main objective of this research was to develop a Bayesian statistical epidemiological model that incorporated social media data and natural language processing to predict COVID-19 cases. Method: This research was developed under the data science paradigm, in which data extracted from social networks (Twitter) and the Influenza Epidemiological Surveillance Information System (SIVEP-Gripe) were used from January/2020 to December/2020. As a theoretical basis for the contextual analysis of social media data, a theoretical model from the field of health communication was used – the Health Belief Model. The techniques used were: (1) natural language processing; (2) social network analysis; and (3) Bayesian epidemiological model. Finally, to analyze the predictive power of the proposed statistical model, two epidemiological models were developed, one that did not use social media data and the other that incorporated this data. Results: As a main result, it was observed that the epidemiological model that incorporated social media data could better predict COVID-19 cases compared to the model that did not use social media data. Another relevant result was the contribution to the health communication literature, where the extension of the Health Belief Model in the context of the COVID-19 pandemic was presented. Conclusion: This research provides relevant information for researchers and other stakeholders in the pandemic response by understanding how social media can be used to improve epidemiological models, enabling the enhancement of prediction estimates, which, in turn, contributes to mitigating risks associated with pandemics. Consequently, it supports policymakers in developing and implementing more effective public health strategies.