From #Jobsearch to #Mask: Improving COVID-19 Cascade Prediction with Spillover Effects

As the pandemic of social media panic spreads faster than the COVID-19 outbreak, an urgent challenge arises: a prediction model needs to be developed to predict the future diffusion size of a piece of COVID-19 information at an early stage of its emergence. In this paper, we focus on the cascade prediction of COVID-19 information with spillover effects. We build the first COVID-19-related Twitter dataset of the Greater Region from the cascade perspective and explore the structure of the cascades. Moreover, the existence of spillover effects is verified in our data and spillover effects for information on COVID-19 symptoms, anti-contagion and treatment measures are found to be from multiple topics of other information. Building on the above findings, we design our SE-CGNN model (CoupledGNN with spillover effects) based on CoupledGNN for cascade prediction. Experiments conducted on our dataset demonstrate that our model outperforms the state-of-the-art methods for COVID-19 information cascade prediction.