News media is often referred to as the Fourth Estate, a recognition of its political power. New understandings of how media shape political beliefs and influence collective behaviors are urgently needed in an era when public opinion polls do not necessarily reflect election results and users influence each other in real-time under algorithm-mediated content personalization. In this work, we measure not only the average but also the distribution of audience political leanings for different media across different countries. The methodological components of these new measurements include a high-fidelity COVID-19 tweet dataset; high-precision user geolocation extraction; and user political leaning estimated from the within-country retweet networks involving local politicians. We focus on geolocated users from eight countries, profile user leaning distribution for each country, and analyze bridging users who have interactions across multiple countries. Except for France and Turkey, we observe consistent bi-modal user leaning distributions in the other six countries, and find that cross-country retweeting behaviors do not oscillate across the partisan divide. More importantly, this study contributes a new set of media bias estimates by averaging the leaning scores of users who share the URLs from media domains. Through two validations, we find that the new average audience leaning scores strongly correlate with existing media bias scores. Lastly, we profile the COVID-19 news consumption by examining the audience leaning distribution for top media in each country, and for selected media across all countries. Those analyses help answer questions such as: Does center media Reuters have a more balanced audience base than partisan media CNN and Fox News in the US? Does far-right media Breitbart attract any left-leaning readers in any countries? Does CNN reach a more balanced audience base in the US than in UK and Spain? In sum, our data-driven methods allow us to study media that are not often collected in editor-curated media bias reporting, especially in non-English-speaking countries. We hope that such cross-country research would inform media outlets of their effectiveness and audience bases in different countries, inform non-government and research organizations about the country-specific media audience profiles, and inform individuals to reflect on our day-to-day media diet.

Proceedings of the ACM on Human-Computer Interaction, Volume 7, Issue CSCW2, Article No.: 317, pp. 1–29

Access the paper here: https://doi.org/10.1145/3610108

— Cai Yang, Lexing Xie, Siqi Wu