Which News Goes Viral? Measuring Identity Threats and Engagement on Social Media

• In Event: What Makes It Go Viral?

Abstract

Prior to the digital era, news outlets had limited knowledge of the content demanded by their audience. But websites and news sharing via social media provide news outlets with granular information about audience engagement. One hypothesis is that news stories which include content about meaningful social identities--including identities related to race/ethnicity, gender, religion, and partisanship--are more likely to be spread and to generate engagement through social media. This pattern, in turn, may encourage media outlets to provide more such identity-oriented coverage over time. We use observational and experimental techniques to test these hypotheses. Observationally, we analyze the full universe of tweets from 20 major media outlets to determine the extent to which identity-oriented news stories posted on social media are more likely to be shared or engaged with. Specifically, using tools from Natural Language Processing including ensemble methods, we build a social identity classifier, and apply it to two novel, large-scale data sets to track the aspects of news stories posted on social media which make them more or less likely to be shared or generate reactions via Twitter (2008-2021) and Facebook (2017-2021). Facebook data is accessed through the Social Science One collaboration, and corresponding statistical corrections to address differential privacy are applied. We focus on several salient social identities, including racial/ethnic identities, gender-based identities, religious identities, and political identities. In addition, we report multiple, large-scale experiments run on Facebook in which we randomly vary the salience of identity-oriented cues present in a news story. Preliminary evidence indicates that the observational and experimental results diverge in instructive ways. While we see clear evidence that news media posts with social identity content are more likely to generate engagement on Twitter and Facebook, heightening the salience of identity cues within a given story does not produce comparable effects.

Authors

• Daniel J. Hopkins, University of Pennsylvania

• Yphtach Lelkes, University of Pennsylvania

• Sam Wolken