Association of specific keywords index in Baidu website with influenza monitoring data during 2012 – 2020 and potential use of the index for influenza epidemic prediction

  Objective  The analyze change trend in the correlation between specific keywords index in Baidu website (keyword Baidu index) and influenza monitoring data in China and to construct a Baidu index-based prediction model for influenza epidemics.

  Methods   The data on weekly number of influenza virus-positive cases in China from the first week of 2012 through the 12th week of 2020 were collected from the Global Influenza Surveillance and Response Network (GISRS). Using influenza-related keywords of four domains (disease name, prevention, treatment, and symptom) screened out with literature studies, the daily Baidu indexes of those keywords during the same period were extracted from the Baidu index database (http://index.Baidu.com/). Using SPSS 22.0 software, the coefficients for the correlation between the keyword Baidu index and the number of influenza virus-positive cases were calculated for the two periods separated by a time node of 2017, when the scale of influenza epidemics in China changed significantly, and multivariate linear regression models for the correlation between the two variables were also constructed with Eviews 8 software.

  Results   Totally 70 keywords were screened out. Before 2017, there were 18 keywords with the coefficients of greater than 0.5 for the correlation between keyword Baidu index and the weekly number of virus-positive influenza cases; while there were 30 such keywords after 2017, among which, 28 keywords were with the coefficients of much greater than 0.5. The top four keywords with the greatest coefficients were swine flu, influenza, influenza A, and fever before 2017; but after 2017, the top four keywords were swine flu symptom, influenza symptom, medicine for the treatment of influenza, and Tylenol (product name for paracetamol). The constructed regression model with the independent variables including a non-specific keyword of ‘high fever’ was of a better prediction outcome, and the prediction deviation of the model was reduced when the specific keyword was replaced.

  Conclusion   In China, the scope of influenza-related keywords has been extending in network media based on big data monitoring and the correlation between the keywords with influenza epidemic has also been increased. The public tend to acquire more information on symptoms and treatment of influenza than on general knowledge about the infectious disease. The results suggest that the selected keywords should be updated timely and more specific keywords should be adopted in influenza epidemic surveillance with big data from network media.