Tagged Twitter timelines for users reporting SARS-CoV-2 infections and related data

Twitter data was collected through the Twitter API v2.0, specifically through the timeline endpoint. Details about the inference of SARS-CoV-2 self-reports and the tagging of the full timeline of each user can be found in the GitHub repository. The larger dataset ("preprocessed_data.csv") consists of a total of 8,534,171 tweets posted by 30,856 users from January 1, 2020 to October to September 30, 2021.

The raw data from Twitter, including tweet and user IDs, has been removed or anonymized in order to comply with the EPFL guidelines for data sharing.

In particular, the date of the tweets was removed, the text of the tweets, URLs and URL domains have been substituted with the "text", "" and "" token respectively.

User IDs have been substitued with new IDs in the [0, number of users] range (e.g. U0, U1, ...) .

Tweet IDs have been substitued with new IDs in the [0, number of tweets] range (e.g. T0, T1, ...) .

In addition to self-explanatory columns about topics, emotions, URL classification and symptoms we tagged, we also share the columns:

pdate: date of the SARS-CoV-2 infection self-report for that user (adjusted with SUTime)

effective_date: date of the tweet adjusted with SUTime, when the SUTime columns is available.

rel_effective_day(week, month): days (weeks, months) computed with respect to the positivity date (i.e. "pdate" column). Negative numbers refer to tweets posted before the user reported a COVID-19 infection on Twitter.