When a Virus Goes Viral: A Study on the Efficacy of Using Twitter Analysis to Forecast COVID-19 Cases

Abstract

In a short period of time, COVID-19 has completely transformed the landscape of global health, economics, and society. Given the enormity of this impact, it has become crucial to more effectively prepare for and act against COVID-19; improving our ability to forecast case counts is one method of doing so. This paper discusses a forecasting model which aims to quantify an aspect of social response in order to build a more well-rounded predictor of case trends. Specifically, by analyzing Twitter data for sentiment and frequency, the model hopes to take into account social attitudes and behaviors towards COVID-19. This data is considered in conjunction with reported COVID-19 case data and state demographic information, inputted into a feedforward neural network model for regression, and ultimately used to forecast positive cases 3, 7, and 14 days into the future.