基于Twitter数据的时间序列模型在流行性感冒预测中的应用

Application of Twitter time series model in influenza prediction

目的为了研究社交媒体数据和流行性感冒(流感)暴发的关系,为有效预防流感提供参考信息。方法利用网络爬虫收集澳大利亚2015年第1周至2017年第29周的Twitter数据,获得社交媒体中的流感相关讨论数据,通过关键字过滤技术获取Twitter中包含sore throat、stuffy nose等有意义推文,采用ARIMA模型和Holt-Winters指数平滑模型对提取的推文列表建模,并进行对比。结果基于Twitter数据的ARIMA模型平均相对误差为6.40%,Holt-Winters指数平滑模型平均相对误差为17.06%。结论基于Twitter数据的ARIMA模型能够有效的预测澳大利亚流感病例的发展趋势。

Objective To discuss the application of Twitter time series model in the prediction of influenza,and to provide evidence for disease prevention and control.Methods Twitter data of first week of 2015 to the 29th week of 2017 in Australia were collected using web crawler and influenza-related discussion data in social media were obtained.Meaningful tweets including key words as sore throat and stuffy nose were screened using keyword filtering technology,and were modeled and compared by using the ARIMA model and the Holt-Winters exponential smoothing model.Results The mean relative error(MRE)was 6.40%for the ARIMA model and was 17.06%for the Holt-Winters exponential smoothing model based on the extracted Twitter data.Conclusions The ARIMA model based on Twitter data can effectively predict the influenza trend in Australia.