Ideology and polarization set the agenda on social media

TL;DR

This study investigates how ideology versus actor prominence shapes online discourse across three global debates (COP26, COVID-19, Ukraine) using large-scale Twitter data and latent-ideology inference. It finds that engagement is driven by ideological alignment, not actor category, with two persistent polarized communities whose stances correlate across topics. Cross-topic consistency is strong (often >90% for majorities) though Ukraine shows weaker cohesion for minorities, challenging simple gatekeeping models. The findings imply decentralized, ideology-driven diffusion of narratives and highlight enduring echo chambers, with implications for outreach strategies and policy design; analyses are English-language and Twitter-only, suggesting avenues for broader cross-platform validation.

Abstract

The abundance of information on social media has reshaped public discussions, shifting attention to the mechanisms that drive online discourse. This study analyzes large-scale Twitter (now X) data from three global debates--Climate Change, COVID-19, and the Russo-Ukrainian War--to investigate the structural dynamics of engagement. Our findings reveal that discussions are not primarily shaped by specific categories of actors, such as media or activists, but by shared ideological alignment. Users consistently form polarized communities, where their ideological stance in one debate predicts their positions in others. This polarization transcends individual topics, reflecting a broader pattern of ideological divides. Furthermore, the influence of individual actors within these communities appears secondary to the reinforcing effects of selective exposure and shared narratives. Overall, our results underscore that ideological alignment, rather than actor prominence, plays a central role in structuring online discourse and shaping the spread of information in polarized environments.

Figures (6)

• Figure 1: Distribution of the number of retweets received by each influencer category and for each debate. The vertical lines reported over each distribution indicate, from left to right, the 0.05, 0.5, and 0.95 quantiles. We also report the usernames of the two most retweeted accounts per debate. No category appears to dominate over the others, but rather singular prominent accounts in each debate emerge over the rest. Notable examples are @GretaThunberg in the climate debate and @KyivIndependent in the debate about the Russo-Ukrainian War.
• Figure 2: Ideological spectra of users and influencers in the COP26, COVID-19, and Ukraine debates. The majority group is mapped to $-1$, whereas the minority group to $+1$. All distributions exhibit bimodality, indicating the presence in each debate of two main communities of users who retweet distinct sets of influencers.
• Figure 3: Joint ideology spectra of users for each pair of debates, with marginal distributions on the top and right sides. The 'Majority' ('Minority') ideology is mapped to the left (right) of the $x$-axis and the bottom (top) of the $y$-axis. Colors indicate the density of users in a region, with lighter colors corresponding to a higher concentration. Most users occupy the regions on the bottom-left and the top-right of the ideology space, indicating a substantial overlap between the Majority and Minority communities across debates.
• Figure 4: Conditional probability of user ideology across debates. Each tile reports the fraction of users in a 'From' debate who belong to the same community in a 'To' debate. More than 90% of users with 'Majority' ideology on a debate (left panel) stay in the majority community of the other two. Similarly, but to a lesser extent, users with 'Minority' ideology on a debate (right panel) tend to stay in the minority community of the other two, especially users within the ideological minority of the COVID-19 debate.
• Figure 5: Median number of retweets and retweeters attracted by influencers of the majority and minority communities. The estimates result from a bootstrapping procedure, with error bars corresponding to the 95% bias-corrected and accelerated confidence intervals. Across all debates, the two sets of influencers receive a comparable number of retweets from a similar number of unique users.