Title:From chambers to echo chambers: Quantifying polarization with a second-neighbor approach applied to Twitter's climate discussion

Abstract:Social media platforms often foster environments where users primarily engage with content that aligns with their existing beliefs, thereby reinforcing their views and limiting exposure to opposing viewpoints. In this paper, we analyze X (formerly Twitter) discussions on climate change throughout 2019, using an unsupervised method centered on chambers--second-order information sources--to uncover ideological patterns at scale. Beyond direct connections, chambers capture shared sources of influence, revealing polarization dynamics efficiently and effectively. Analyzing retweet patterns, we identify echo chambers of climate believers and skeptics, revealing strong chamber overlap within ideological groups and minimal overlap between them, resulting in a robust bimodal structure that characterizes polarization. Our method enables us to infer the stance of high-impact users based on their audience's chamber alignment, allowing for the classification of over half the retweeting population with minimal cross-group interaction, in what we term augmented echo chamber classification. We benchmark our approach against manual labeling and a state-of-the-art latent ideology model, finding comparable performance but with nearly four times greater coverage. Moreover, we find that echo chamber structures remain stable over time, even as their members change significantly, suggesting that these structures are a persistent and emergent property of the system. Notably, polarization decreases and climate skepticism rises during the #FridaysForFuture strikes in September 2019. This chamber-based analysis offers valuable insights into the persistence and fluidity of ideological polarization on social media.