“…The study empirically investigates the effect of sentiments towards COVID-19 predictions in the early days. It does not extend to a later period as (i) a deluge of Tweets mentioned COVID-19 as it became a common topic, which makes the retrieval of data (limited by the company twitter) and the processing extremely slow; and (ii) there is a selection problem because the later Tweets often mentioned COVID-19 casually rather than talking about it, however, in NLP, topic modelling remains an active research topic [65] , [66] . Besides the technical issues in obtaining sentiment indicators, generalization and adaptation of the proposed method in pandemic prediction concern several scales, such as adaptation across events [67] , adaptation to different stages within an event, and adaptation to countries or cities [68] , [69] .…”