Social media provides an essential platform for shaping and sharing opinions and consuming information in a decentralized way. However, users often interact with and are exposed to information mostly aligned with their beliefs, creating a positive feedback mechanism that reinforces those beliefs and excludes contrasting ones. In this paper, we study such mechanisms by analyzing the social network dynamics of controversial Twitter discussions using unsupervised methods that de- mand little computational power. Specifically, we focus on the retweet networks of the climate change conversation during 2019, when important climate social movements flourished. We find echo chambers of climate believers and climate skeptics that we identify based solely on the retweeting patterns of Twitter users. In particular, we study the information sources, or chambers, consumed by the audience of the leading users of the conversation. Users with similar (contrasting) ideological positions show significantly high (low)-overlapping chambers, resulting in a bimodal overlap distribution. Further, we uncover the ideological position of previously unobserved high-impact users based on how many audience mem- bers fall into either echo chamber. We uncover the ideology of more than half of the retweeting population as either climate believers or skeptics and find that the cross-group communication is small. Moreover, we find that, while the echo chamber structures are consistent throughout the year, most users inside the echo chambers change from one week to the next, suggesting that they are a stable emergent property of the Twittersphere. Interestingly, we detect a significant de- crease in polarization and a significant increase on the number of climate skeptics that coincides with the main #FridaysForFuture strikes during September, 2019.


Kolic, B., Aguirre-López, F., Hernández-Williams, S. & Garduño-Hernández, G. (2022). 'Quantifying the structure of controversial discussions with unsupervised methods: a look into the Twitter climate change conversation'. INET Oxford Working Paper No. 2022-23.
Download Document (pdf, 5.764 MB)