关键词: 时空地理加权回归模型, 搜索引擎数据, 流感, 空间分布模型

Abstract:

The Internet records people's daily life,and the analyzing and mining of search engine query data with location can discover valuable geographic information hidden in it.In this paper,the correlation between the monthly influenza case data in each Chinese province and the Baidu search index of related keywords were calculated,the most related keyword's index was chosen as the explanatory variable while the influenza case data was chosen as the dependent variable.The principal component analysis was used to eliminate the effect of multicollinearity among variables before the spatial distribution model of influenza was constructed by ordinary least squares regression (OLS),geographically weighted regression (GWR) and geographically and temporally weighted regression (GTWR).The GTWR model demonstrated a better goodness-of-fit (0.959) than the OLS (0.737) and GWR model (0.915).The Akaike information criterion (AIC) test also supported that the improvement made by GTWR over OLS and GWR models were statistically significant.Validation results showed that the GTWR model can accurately identify the high prevalence area of influenza.It demonstrates that combining the GTWR model with search engine query data can model the spatial distribution of influenza accurately,and provide a prediction model and statistical explanation for the study of epidemiology.

Key words: geographically and temporally weighted regression, search engine data, influenza, spatial distribution model