This study explores the internal mechanism and possibility of forecasting an influenza epidemic based on both search queries and actual influenza data. First, the logical relationship is explored between online information searches and conventional surveillance data based on the concepts of information behaviors, information seeking be-haviors, and so on. Then, the range selection method and cross-correlation analysis are used to select keywords ac-cording to the theoretical framework. Finally, three models are established and compared. The results show that (i) the empirical research proves the logical rationality of the theoretical framework: the keywords that could reflect flu trends ten weeks in advance are related to influenza vaccines; those a week in advance are related to influenza symp-toms; and most of the simultaneous keywords are frequent terms related to influenza; (ii) all three models can predict influenza effectively, and support vector machine yields the most accurate forecasting result.%本文通过挖掘网络搜索数据与我国流感疫情的在内在机理,利用关键词的时序特征实现了较为精准的提前预测.研究首先从信息行为、信息搜寻行为等理论概念出发,对百度指数与流感病例数据之间的逻辑关系进行探讨,建立理论框架;然后以理论框架为基础,用范围选词法对百度搜索词进行初步筛选,并利用互相关分析选出具有先行性质的关键词,用于构建预测模型;最后,对比融合百度指数的三种预测模型,评估其预测效果.互相关分析结果大致符合本文提出的逻辑框架,可提前十周预测流感疫情的关键词内容和流感疫苗相关;提前一周的关键词多涉及流感的症状表现;而同步类关键词多为常用搜索词或治疗方法.模型对比结果显示,多元线性回归模型、支持向量机模型和神经网络模型都能有效地进行流感预测,无论提前十周还是提前一周,支持向量机的效果最好.
展开▼
