Galbraith Elroy, Li Jie, Rio-Vilas Victor J Del, Convertino Matteo
Nexus Group, Faculty and Graduate School of Information Science and Technology, Hokkaido University, Sapporo, Japan.
Informatics Institute, University of Amsterdam, Amsterdam, The Netherlands.
Sci Rep. 2022 Apr 6;12(1):5831. doi: 10.1038/s41598-022-09656-1.
Social media can forecast disease dynamics, but infoveillance remains focused on infection spread, with little consideration of media content reliability and its relationship to behavior-driven epidemiological outcomes. Sentiment-encoded social media indicators have been poorly developed for expressed text to forecast healthcare pressure and infer population risk-perception patterns. Here we introduce Infodemic Tomography (InTo) as the first web-based interactive infoveillance cybertechnology that forecasts and visualizes spatio-temporal sentiments and healthcare pressure as a function of social media positivity (i.e., Twitter here), considering both epidemic information and potential misinformation. Information spread is measured on volume and retweets, and the Value of Misinformation (VoMi) is introduced as the impact on forecast accuracy where misinformation has the highest dissimilarity in information dynamics. We validated InTo for COVID-19 in New Delhi and Mumbai by inferring distinct socio-epidemiological risk-perception patterns. We forecast weekly hospitalization and cases using ARIMA models and interpolate spatial hospitalization using geostatistical kriging on inferred risk perception curves between tweet positivity and epidemiological outcomes. Geospatial tweet positivity tracks accurately [Formula: see text]60[Formula: see text] of hospitalizations and forecasts hospitalization risk hotspots along risk aversion gradients. VoMi is higher for risk-prone areas and time periods, where misinformation has the highest non-linear predictability, with high incidence and positivity manifesting popularity-seeking social dynamics. Hospitalization gradients, VoMi, effective healthcare pressure and spatial model-data gaps can be used to predict hospitalization fluxes, misinformation, healthcare capacity gaps and surveillance uncertainty. Thus, InTo is a participatory instrument to better prepare and respond to public health crises by extracting and combining salient epidemiological and social surveillance at any desired space-time scale.
社交媒体可以预测疾病动态，但信息监测仍主要聚焦于感染传播，很少考虑媒体内容的可靠性及其与行为驱动的流行病学结果之间的关系。用于预测医疗压力和推断人群风险感知模式的情感编码社交媒体指标在表达文本方面发展不足。在此，我们引入信息疫情断层扫描（InTo），这是首个基于网络的交互式信息监测网络技术，它能预测并可视化时空情感以及作为社交媒体积极性（即此处的推特）函数的医疗压力，同时考虑疫情信息和潜在的错误信息。信息传播通过数量和转发量来衡量，并引入错误信息价值（VoMi）来衡量错误信息在信息动态方面具有最高差异时对预测准确性的影响。我们通过推断不同的社会流行病学风险感知模式，在新德里和孟买对InTo针对新冠疫情进行了验证。我们使用自回归积分移动平均（ARIMA）模型预测每周的住院人数和病例数，并利用地理统计克里金法在推文积极性与流行病学结果之间推断出的风险感知曲线上对空间住院情况进行插值。地理空间推文积极性准确跟踪了约60%的住院情况，并沿着风险规避梯度预测住院风险热点。在易发生风险的地区和时间段，VoMi较高，在这些地方错误信息具有最高的非线性可预测性，高发病率和高积极性表现出追求人气的社会动态。住院梯度、VoMi、有效医疗压力以及空间模型 - 数据差距可用于预测住院流量、错误信息、医疗能力差距和监测不确定性。因此，InTo是一种参与性工具，通过在任何期望的时空尺度上提取并结合显著的流行病学和社会监测信息，更好地应对公共卫生危机。
