Tracking COVID-19 using online search

arXiv 2020  ·  Vasileios Lampos, Simon Moura, Elad Yom-Tov, Michael Edelstein, Maimuna Majumder, Yohhei Hamada, Molebogeng X. Rangaka, Rachel A. McKendry, Ingemar J. Cox ·

Research outcomes over the years have showcased the capacity of online search behaviour to model various properties of infectious diseases. In this work we use online search query frequency time series to gain insights about the prevalence of COVID-19 in multiple countries. We first develop unsupervised modelling techniques based on identified symptom categories by United Kingdom's National Health Service. We then propose ways for minimising an expected bias in these signals partially generated by the early and continuous exposure to news media. We also look into transfer learning techniques for mapping supervised models from countries where the disease spread has progressed to countries that are in earlier phases of the epidemic curve. Furthermore, we analyse the time series of online search queries in relation to confirmed COVID-19 cases data jointly across multiple countries, uncovering interesting patterns. Finally, we show results from short-term forecasting models based on Gaussian Processes that combine confirmed cases and online search data time series.

PDF Abstract

Datasets


  Add Datasets introduced or used in this paper