Source-linked AI summary
Political Discourse on Social Media: Echo Chambers, Gatekeepers, and the Price of Bipartisanship
Kiran Garimella, Gianmarco De Francisci Morales, Aristides Gionis, Michael Mathioudakis
TL;DR
Political echo chambers may limit exposure to opposing views, but evidence has not consistently combined shared content with the networks through which it propagates. This paper measures users’ political content production and consumption across Twitter datasets and examines partisan, bipartisan, and gatekeeper roles. It finds prevalent political echo chambers, costs for users producing cross-partisan content, and greater difficulty identifying gatekeepers than partisans.
Problem
The paper addresses how political echo chambers exist and are structured on social media, where users may be surrounded by opinions agreeing with their own.
Method
The authors jointly measure the political leaning of content users share and receive and relate these measures to the social network around each user.
Results
Political echo chambers are prevalent on Twitter; bipartisan users face costs in centrality, community connection, and endorsements, while gatekeepers are harder to predict than partisan users.
Takeaways & Limitations
Understanding echo chambers requires considering user attitudes and the interdependence of content production, consumption, and network properties.
Takeaways & Limitations
The observations come from politically savvy users on Twitter and should not be assumed to generalize immediately to other settings or the general public.
Abstract
from arXiv · showhide
Echo chambers, i.e., situations where one is exposed only to opinions that agree with their own, are an increasing concern for the political discourse in many democratic countries. This paper studies the phenomenon of political echo chambers on social media. We identify the two components in the phenomenon: the opinion that is shared ('echo'), and the place that allows its exposure ('chamber' --- the social network), and examine closely at how these two components interact. We define a production and consumption measure for social-media users, which captures the political leaning of the content shared and received by them. By comparing the two, we find that Twitter users are, to a large degree, exposed to political opinions that agree with their own. We also find that users who try to bridge the echo chambers, by sharing content with diverse leaning, have to pay a 'price of bipartisanship' in terms of their network centrality and content appreciation. In addition, we study the role of 'gatekeepers', users who consume content with diverse leaning but produce partisan content (with a single-sided leaning), in the formation of echo chambers. Finally, we apply these findings to the task of predicting partisans and gatekeepers from social and content features. While partisan users turn out relatively easy to identify, gatekeepers prove to be more challenging.
1 INTRODUCTION
The paper studies political echo chambers on Twitter by jointly examining the political leaning of shared and received content and the social network enabling its propagation. It finds prevalent agreement between users’ produced and consumed political content, while bipartisanship carries network and engagement costs.
- Echo chambers are concerning because citizens may hear mainly opinions aligned with their own, potentially hampering democratic deliberation.
- The study defines an echo chamber as agreement between the political leaning of content users share and content they receive from their network.
- Production and consumption measures applied to Twitter datasets reveal a large correlation between the leaning of content users produce and consume.The datasets include one collection of over 2.5 billion tweets spanning almost eight years.
- Bipartisan users pay a price in network centrality, community connection, and content endorsements such as retweets and favorites.
- Gatekeepers consume content from both political leanings but produce single-sided content, occupying an intermediate network position while aligning with one side.They are described as a small group with higher-than-average network centrality and limited community embeddedness.
- Partisan users are relatively easy to identify from social and content features, whereas gatekeepers are more challenging to predict.
- The study establishes political echo chambers on Twitter while motivating more nuanced analyses of user attitudes and the interaction between content and network structure.
2 RELATED WORK
Prior work studies echo chambers through content, networks, or interactions, but lacks a consistent definition and often treats consumption separately from production. This paper jointly characterizes content and network properties and introduces a paper-specific definition of gatekeepers.
- Echo chambers describe situations in which users consume content expressing the same viewpoint they hold or express.
- Earlier studies examined agreement in blogs, cross-cutting content on Facebook, and political polarization in Twitter interaction networks.
- Facebook research distinguishes potential exposure, feed exposure, and engagement, finding that users engage with less cross-cutting content despite significant exposure.
- Because the literature lacks a consistent definition, this paper measures echo chambers using users’ jointly analyzed content production, consumption, and network position.
- Selective exposure, biased assimilation, algorithmic filtering, and personalization are related mechanisms, but their interactions and causal relations are outside the paper’s scope.
- The paper’s price-of-bipartisanship analysis links producing both political leanings with costs in network centrality and content engagement.
- The paper defines gatekeepers as users who receive content from both political leanings but produce content from only one, thereby filtering information from one side.
3 DATA
The study combines ten Twitter datasets covering political and non-political topics with directed follow networks and source-based political leaning scores. The data range from event-specific collections to a politically active user corpus exceeding 2.5 billion tweets.
- The collection contains ten Twitter datasets divided into five politically contentious and five non-political topics.
- For each dataset, the authors construct a directed follow graph in which an edge u →v means user u follows user v.
- Political datasets include guncontrol, obamacare, abortion, a combined election-related collection, and a large longitudinal corpus.
- Political event datasets collect tweets during a one-week window around relevant events and focus on users actively engaged in the discussion.
- The large dataset contains over 2.5 billion tweets from politically active users spanning almost eight years from 2009 to 2016.
- Political leaning is estimated using source-polarity scores for news domains, including scores for 500 widely shared domains ranging from 0 to 1.
4 MEASURES
The paper measures political leaning in users’ produced and consumed content, then relates those measures to partisan status, network position, and content appreciation. It also defines gatekeepers as users whose consumption spans both sides while production remains one-sided.
- Content measures: Production polarity averages the political leaning of tweets a user posts, with values near 0 indicating liberal sources and near 1 indicating conservative sources.The measure uses tweets linking to news organizations with known political leanings.
- Content measures: δ-partisan users have production polarity within δ of either extreme, and smaller δ identifies more strongly partisan users.Users outside this criterion are δ-bipartisan; Figure 1 illustrates the threshold regions.
- Content measures: Production and consumption variance measure the range of political leanings represented in a user’s posted and received content.Consumption variance specifically quantifies the range of opinions covered by consumed content.
- Content measures: Consumption polarity averages the political leaning of tweets a user receives from accounts they follow.Values near 0 indicate liberal consumption, while values near 1 indicate conservative consumption.
- Content measures: Gatekeepers consume content from both political sides but produce content from only one, formally combining non-δ-consumer status with δ-partisan status.The definition treats these users as filters that block or filter information from one side.
- Network measures: Network measures capture users’ positions and interactions through user polarity, PageRank centrality, clustering coefficient, and retweet/favorite rates and volumes.PageRank reflects network importance, clustering reflects embedding in a tightly knit community, and rates differ from volumes by measuring acceptance rather than popularity.
5 ANALYSIS
The analysis tests echo chambers, partisan advantages, gatekeeper characteristics, and predictability using polarity, network, interaction, profile, and content features. Political production and consumption polarities align strongly, while partisans generally show greater polarization, centrality, connectivity, and appreciation than bipartisans.
- Echo chambers: Production and consumption polarities are examined jointly to test whether users receive content matching their political leaning or opposite-leaning content.The analysis uses joint polarity distributions and source-polarity distributions across political and non-political datasets.
- Echo chambers: Production and consumption polarities are highly correlated in Political datasets, whereas their distributions largely coincide in Non-Political datasets.Political datasets show separated, bimodal distributions for Democrats and Republicans; Non-Political distributions largely overlap.
- Echo chambers: Bipartisan users follow sources with a wider spread of political leanings, while partisan users show lower polarity variance and more concentrated production and consumption.Variance follows a downward-U relationship with mean polarity, supporting narrower source ranges among more partisan users.
- Partisan advantages: Partisan users have significantly greater polarization, higher PageRank, higher clustering coefficient, and more appreciated tweets than bipartisan users.The trends remain consistent across δ thresholds, although the appreciation effect is smaller and profile features show no consistent differences.
- Gatekeepers: Gatekeepers have above-average PageRank and in-degree but lower clustering coefficients and less extreme polarity than the rest of their aligned group.Their cross-community links are less likely to form triangles, consistent with consuming content from the opposing side.
- Gatekeepers: Partisans have higher PageRank than gatekeepers, with the difference becoming more pronounced at higher δ thresholds.The authors suggest this pattern may indicate that not following users from the opposite side is rewarded among polarized users.
- Prediction: Using combined network and content features, partisan prediction reaches about 80% accuracy, compared with about 70% for gatekeeper prediction.Partisans are relatively easy to identify, whereas gatekeepers are more challenging; the content features use tweets linking to news sources rather than article text.
6 DISCUSSION
The discussion characterizes political echo chambers through the interplay of network position and content leaning, while identifying distinct user roles and important empirical boundaries. It also highlights limits on generalization and measurement, alongside directions for richer future analysis.
- Characterising echo chambers: Production and consumption measures capture the interplay between users’ shared content and the content reaching them through their social-network neighborhoods.
- Characterising echo chambers: Political echo chambers appear in contentious discussions: production and consumption polarities are bimodal and highly correlated, but the pattern is absent for non-contentious topics.
- Partisan users: Bipartisan users pay a price in network centrality, community connections, and endorsements such as retweets and favorites.
- Gatekeepers: Gatekeepers consume both political leanings but produce one-sided content, occupy positions between opposed communities, and usually have lower clustering because their cross-community links remain open.
- Gatekeepers: An alternative gatekeeper definition produced almost identical results, including users with high consumption variance and low production variance.
- Limitations and future work: The findings are limited by Twitter-focused samples, external news-source labels, a follow-based consumption proxy, and a small fraction of potentially non-political articles.