An augmented multilingual Twitter dataset for studying the COVID-19 infodemic

Abstract

This work presents an openly available dataset to facilitate researchers' exploration and hypothesis testing about the social discourse of the COVID-19 pandemic. The dataset currently consists of over 2.2 billions tweets (count as of September, 2021), from all over the world, in multiple languages. Tweets start from January 22, 2020, when the total cases of reported COVID-19 were below 600 worldwide. The dataset was collected using the Twitter API and by rehydrating tweets from other available datasets, data collection is ongoing as of the time of writing. To facilitate hypothesis testing and exploration of social discourse, the English and Spanish tweets have been augmented with state-of-the-art Twitter Sentiment and Named Entity Recognition algorithms. The dataset and the summary files provided allow researchers to avoid some computationally intensive analyses, facilitating more widespread use of social media data to gain insights on issues such as (mis)information diffusion, semantic networks, sentiments, and the evolution of COVID-19 discussions. In addition, the dataset provides an archive for researchers in the social sciences wishing to have access to a dataset covering the entire duration of the pandemic.

Keywords: COVID-19; Named Entity Recognition; Sentiment analysis; Twitter.

Fig. 1

Example of Tweet Related to COVID-19

Fig. 2

Example of dataset tables

Fig. 3

Tweet frequency across top five observed languages

Fig. 4

Map of tweets featuring geolocation information

Fig. 5

Sentiment of English-language tweets

Fig. 6

Daily proportion of English-language tweets sentiment

Fig. 7

Sentiment of Spanish tweets

Fig. 8

Daily proportion of Spanish-language tweets by sentiment

Fig. 9

Network generated from English tweets augmented dataset