Enhancing the Predictive Power of Google Trends Data Through Network Analysis: Infodemiology Study of COVID-19

Abstract

Background The COVID-19 outbreak has revealed a high demand for timely surveillance of pandemic developments. Google Trends (GT), which provides freely available search volume data, has been proven to be a reliable forecast and nowcast measure for public health issues. Previous studies have tended to use relative search volumes from GT directly to analyze associations and predict the progression of pandemic. However, GT’s normalization of the search volumes data and data retrieval restrictions affect the data resolution in reflecting the actual search behaviors, thus limiting the potential for using GT data to predict disease outbreaks. Objective This study aimed to introduce a merged algorithm that helps recover the resolution and accuracy of the search volume data extracted from GT over long observation periods. In addition, this study also aimed to demonstrate the extended application of merged search volumes (MSVs) in combination of network analysis, via tracking the COVID-19 pandemic risk. Methods We collected relative search volumes from GT and transformed them into MSVs using our proposed merged algorithm. The MSVs of the selected coronavirus-related keywords were compiled using the rolling window method. The correlations between the MSVs were calculated to form a dynamic network. The network statistics, including network density and the global clustering coefficients between the MSVs, were also calculated. Results Our research findings suggested that although GT restricts the search data retrieval into weekly data points over a long period, our proposed approach could recover the daily search volume over the same investigation period to facilitate subsequent research analyses. In addition, the dynamic time warping diagrams show that the dynamic networks were capable of predicting the COVID-19 pandemic trends, in terms of the number of COVID-19 confirmed cases and severity risk scores. Conclusions The innovative method for handling GT search data and the application of MSVs and network analysis to broaden the potential for GT data are useful for predicting the pandemic risk. Further investigation of the GT dynamic network can focus on noncommunicable diseases, health-related behaviors, and misinformation on the internet.

Infodemiology of Influenza-like Illness: Utilizing Google Trends’ Big Data for Epidemic Surveillance

“…As for applying GT in disease prediction, Zhang et al [43] found that temperature-related internet search indicators can predict influenza outbreaks and should be regarded as an effective indicator for modeling influenza outbreaks and monitoring and early warning systems; Ghosh et al [1] also used relevant news trends to monitor the time-varying incidence of diseases. In addition to being used for medical research, GT can be used in other fields, such as explaining economic activities, energy use trends [11,58], agricultural economic development [14,59], and forecasting the unemployment rate [17,26,60,61], stock market changes [24,25,62,63], and tourism behavior [64][65][66][67].…”

“…GT is a service based on big data that provides real-time trend predictions through the frequency of the 'search keywords' ability [11]. Nowadays, many research fields use real-time trends from GT data for analysis and prediction [1,5,[12][13][14]. Due to the sensitivity and immediacy brought by search engine data, Eysenbach [15] pointed out that the data obtained by using Google search engine analysis can provide real-time and accurate predictions and reduce the cost of the entire flu season.…”