Misleading Tweets and Helpful Notes: Investigating Data Labor by Twitter Birdwatch Users

Isaiah Jones, Brent J. Hecht, Nicholas Vincent

Abstract: In response to concerns about misleading content on social media, Twitter launched the “Birdwatch” initiative that allows volunteers to label and add context to tweets. We study data from Birdwatch to understand how users are performing “data labor” for Twitter, with implications for other platforms that are similarly reliant on data labor. We conduct computational analyses of Birdwatch text data and perform machine learning experiments to see how Birdwatch contributions might be used for classification. We find that Birdwatch users discuss distinct topics in domains like politics and news. While using Birdwatch data for content-only predictions may provide only a small amount of predictive power, in some cases Birdwatch data may be able to support ML systems. Furthermore, we see that the continuous flow of Birdwatch contributions provides great value in terms of supporting a “guess most frequent“ baseline for classifying Twitter content.