%0 Journal Article
%9 ACL : Articles dans des revues avec comité de lecture répertoriées par l'AERES
%A Jourdain, F.
%A Chakraborty, Debapriyo
%A Gaillard, Béatrice
%A Gautier, A.
%A Simard, Frédéric
%A Robert, P. J.
%A Dormont, L.
%A Desenclos, J. C.
%A Roche, Benjamin
%T Inferring human behavior through online social networks may provide accurate behavioral estimates for outbreak forecasting of arboviruses
%D 2025
%L fdi:010094407
%G ENG
%J PLoS Global Public Health
%K FRANCE
%M ISI:001536582100005
%N 7
%P e0004889 [12 ]
%R 10.1371/journal.pgph.0004889
%U https://www.documentation.ird.fr/hor/fdi:010094407
%> https://horizon.documentation.ird.fr/exl-doc/pleins_textes/2025-09/010094407.pdf
%V 5
%W Horizon (IRD)
%X Human behavior is known to be a fundamental, yet often neglected, component of infectious disease epidemiology, especially during outbreaks. To quantify its role and fluctuations, analyzing message contents on popular online social networks - part of so-called digital epidemiology - is a promising approach. However, such methods could be biased and generate estimation errors since social media users may not be representative of the general population. To address this, we systematically compared social media-derived estimates with those obtained from a large-scale opinion survey. In the setting of metropolitan France, where the risk of arbovirus outbreaks is increasingly important, we compared the frequency of three types of emotional states related to human-mosquito contact identified in 160,000 messages on X (formerly Twitter) with the frequency of the same emotional states expressed through a large-scale opinion survey involving 15,000 people during the same period. Both sources of data were used to parametrize a mathematical model of mosquito-borne virus transmission. We found that estimates of these emotional states for different age groups in the opinion survey could be highly different from estimates based on X data. Nevertheless, by integrating demographic adjustments and incorporating variability into our transmission models, we showed that the predicted overall outbreak dynamics remain comparable under certain conditions. This study provides the first evidence that using digital social network data to infer epidemiologically relevant behavior achieves similar results as using large-scale opinion survey data. These outcomes highlight that X data could be used to help forecast outbreaks dynamics, opening new opportunities for real-time assessment of human health-related behavior and the definition of control strategies.
%$ 052 ; 056