Ideology and polarization set the agenda on social media

Basic Information

• Paper ID: 2412.05176
• Title: Ideology and polarization set the agenda on social media
• Authors: Edoardo Loru, Alessandro Galeazzi, Anita Bonetti, Emanuele Sangiorgio, Niccolò Di Marco, Matteo Cinelli, Max Falkenberg, Andrea Baronchelli, Walter Quattrociocchi
• Classification: cs.SI (Social and Information Networks), cs.CY (Computers and Society), physics.soc-ph (Physics of Society)
• Published Journal: Nature Scientific Reports (DOI: 10.1038/s41598-025-19776-z)
• Paper Link: https://arxiv.org/abs/2412.05176

Abstract

The abundance of information on social media has reshaped public discourse, directing attention toward the mechanisms driving online conversation. This study analyzes large-scale Twitter data from three global debates (climate change, COVID-19, and the Russia-Ukraine war) to investigate the structural dynamics of participation. The research reveals that discussions are shaped not primarily by specific categories of actors (such as media or activists), but rather by shared ideological alignment. Users consistently form polarized communities, with their ideological positions in one debate predicting their positions in other debates. This polarization transcends individual topics, reflecting broader patterns of ideological division.

Research Background and Motivation

Problem Definition

This research addresses a fundamental question about the structural transformation of public discourse in the social media era: in an environment of information overload, what factors truly drive user engagement and attention allocation? Traditional agenda-setting theory emphasizes the hierarchical influence of institutional actors (such as mainstream media and political elites), but does this model remain applicable in the decentralized environment of social media?

Research Significance

1. Theoretical Challenge: Traditional agenda-setting theory faces challenges from digital transformation and requires reassessment of its applicability in networked public spheres
2. Social Impact: Understanding polarization mechanisms on social media is crucial for evaluating their effects on democratic discourse and social cohesion
3. Practical Value: Provides communication professionals and policymakers with effective communication strategies tailored to ideological communities

Limitations of Existing Approaches

• Traditional research primarily focuses on polarization within specific topics, lacking systematic cross-topic analysis
• Lacks quantitative comparison of the relative importance of actor categories versus ideological alignment
• Absence of large-scale, multi-topic empirical research to validate theoretical hypotheses

Core Contributions

1. Demonstrating the Dominant Role of Ideological Alignment: Proves that ideological alignment rather than actor categories is the primary driver of social media engagement
2. Revealing Cross-Topic Polarization Patterns: First systematic evidence that users maintain high ideological consistency across different topics
3. Challenging Traditional Agenda-Setting Theory: Provides empirical evidence of decentralized, bottom-up discourse dynamics in social media environments
4. Constructing Large-Scale Multi-Topic Dataset: Integrates Twitter data from three major global topics, providing an important foundation for subsequent research

Methodology Details

Task Definition

Research tasks include:

• Input: Twitter data from three topics: COP26, COVID-19, and Russia-Ukraine war
• Output: User ideological distribution, influencer classification, cross-topic ideological consistency analysis
• Constraints: Analysis limited to English-language tweets, focusing on retweet behavior as an engagement indicator

Data Collection and Processing

Data Scale

• COP26: Approximately 8 million English-language tweets, 1 million users (June 1 - November 14, 2021)
• COVID-19: Approximately 35 million English-language tweets, 8 million users (January 1, 2020 - April 30, 2021)
• Ukraine: Approximately 85 million English-language tweets, 7 million users (February 22, 2022 - February 17, 2023)

Influencer Identification Method

1. Initial Screening: Selection of accounts in the top 1% by retweet count for each topic
2. Content Production Filtering: Selection of accounts with original tweet production in the top 50%
3. Manual Annotation: Classification of influencers into six categories (activists, international organizations/NGOs, media, political, individual, other)

Latent Ideology Analysis Method

Technical Principles

Employs the "latent ideology" method, inferring ideological positions based on user retweet interaction patterns with influencers:

1. Interaction Matrix Construction: $M_{ij}$ represents the number of times user $i$ retweets influencer $j$
2. Correspondence Analysis: Singular value decomposition through standardized residual matrix
3. Ideology Scoring: Maps users to an ideological spectrum in the range $[-1, +1]$

Validation Methods

• Hartigan's dip test to verify bimodality of distributions
• Manual annotation to verify accuracy of ideological groupings

Technical Innovations

1. Multi-Topic Comparison Framework: First systematic comparison of three different types of global topics within a single analytical framework
2. Quantitative Comparison of Actor Categories vs. Ideology: Direct statistical comparison of the effectiveness of two explanatory mechanisms
3. Cross-Topic Ideological Persistence Analysis: Innovative quantification of user ideological consistency across different topics

Experimental Setup

Evaluation Metrics

• Kolmogorov-Smirnov Test: Comparison of retweet distributions across different actor categories
• Hartigan's Dip Statistic: Testing bimodality of ideological distributions
• Gini Index: Quantifying inequality in user preferences across actor categories
• Conditional Probability: Calculating cross-topic ideological consistency

Comparison Methods

• Random Permutation: Random reassignment of actor category labels
• Uniform Sampling: Random sampling of category labels from uniform distribution
• Bootstrap Method: Estimating engagement differences among influencers in different ideological communities

Experimental Results

Main Findings

1. Actor Categories Do Not Determine Engagement

• No significant differences in retweet distribution across all actor categories (Kolmogorov-Smirnov test, p > 0.05)
• Observed retweet patterns are consistent with results from random category label assignment
• Individual prominent accounts (such as @GretaThunberg, @KyivIndependent) exert influence beyond category effects

2. Ubiquitous Ideological Bipolarity

All three topics exhibit significant bimodal ideological distributions:

• COP26: D = 0.021 (P < 0.001)
• COVID-19: D = 0.035 (P < 0.001)
• Ukraine: D = 0.068 (P < 0.001)

3. Cross-Topic Ideological Consistency

Probability of users maintaining the same ideological position across different topics:

• Mainstream Population: Over 90% of users maintain consistent positions across topics
• Minority Population: Approximately 70-90% of users maintain consistency, with minority populations in COVID-19 showing the highest consistency (approximately 90%)

Ablation Studies

User Preference Analysis

• Individual users tend to focus on a small number of actor categories (median Gini index: 0.69-0.77)
• However, at the topic level, different categories receive relatively balanced overall attention (Gini index: 0.28-0.59)

Behavioral Differences in Ideological Communities

Different ideological communities exhibit variations in actor category preferences:

• COP26: Minority populations prefer individual accounts (35%), mainstream populations prefer activists (21%) and organizations (28%)
• Ukraine: Both populations prefer media and individual accounts, but differ in preferences for political accounts
• COVID-19: Both ideological communities show similar preferences

Related Work

Evolution of Agenda-Setting Theory

• Classical Theory: McCombs and Shaw's top-down agenda-setting model
• Digital Adaptation: New concepts such as networked agenda-setting and agenda convergence
• Social Media Research: Platform-specific agenda-setting mechanism studies

Social Media Polarization Research

• Echo Chamber Effect: Formation mechanisms of selective exposure and information cocoons
• Ideology Inference: Methods for identifying political leanings based on social network behavior
• Cross-Platform Research: Comparative studies of polarization patterns across different social media platforms

Conclusions and Discussion

Main Conclusions

1. Decentralized Discourse Structure: Public discussions on social media are no longer dominated by traditional institutional gatekeepers, but rather driven by ideological alignment
2. Cross-Topic Polarization Patterns: User ideological positions exhibit persistence across topics, forming structural divisions that transcend specific issues
3. Bottom-Up Agenda-Setting: Traditional hierarchical agenda-setting models are replaced by decentralized, collaborative network processes

Limitations

1. Platform Limitations: Research focuses only on Twitter, potentially not reflecting user dynamics on other platforms
2. Language Limitations: Analysis limited to English-language tweets, potentially missing localized perspectives and region-specific discourse patterns
3. Methodological Simplification: Complex user beliefs are simplified into binary mainstream/minority classifications
4. Temporal Scope: Insufficient exploration of temporal evolution of ideological consistency

Future Directions

1. Cross-Platform Analysis: Extension to comparative research across multiple social media platforms
2. Multilingual Research: Inclusion of global analysis across different languages and cultural backgrounds
3. Temporal Dynamics: Investigation of formation and evolution processes of ideological consistency
4. Intervention Strategies: Design and testing of interventions promoting cross-ideological dialogue

In-Depth Evaluation

Strengths

1. Methodological Innovation: First systematic comparison of the relative importance of actor categories versus ideological alignment
2. Data Scale and Quality: Utilizes three large-scale, high-quality datasets covering global topics of different natures
3. Statistical Rigor: Employs multiple statistical methods to verify robustness of conclusions
4. Theoretical Contribution: Presents important challenges to traditional agenda-setting theory, advancing communication theory in the digital age

Weaknesses

1. Causal Relationships: Research primarily based on correlation analysis, failing to establish clear causal relationships
2. Mechanism Explanation: Relatively weak explanation of the formation mechanisms of ideological alignment
3. Practical Guidance: While communication strategy recommendations are proposed, specific implementation guidance is lacking
4. External Validity: Applicability of research conclusions in non-English environments and other cultural contexts remains to be verified

Impact

1. Academic Value: Provides important theoretical and empirical contributions to social media research, political communication, and computational social science
2. Practical Significance: Offers valuable insights for media professionals, policymakers, and platform designers
3. Methodological Contribution: Provides a replicable framework for large-scale social media data analysis

Applicable Scenarios

1. Political Communication Research: Analysis of opinion polarization patterns in elections and policy debates
2. Crisis Communication Management: Understanding information dissemination dynamics in public health crises and natural disasters
3. Platform Governance: Reference for social media platform algorithm design and content management
4. Public Policy Development: Scientific basis for policy interventions promoting rational public discourse

References

This research cites 78 important references spanning multiple fields including agenda-setting theory, social media polarization, and network science. Particularly noteworthy are McCombs and Shaw's classical agenda-setting theory, Barberá et al.'s latent ideology method, and recent important research on social media echo chamber effects.

Summary: This is a high-quality paper with significant theoretical and empirical value in the field of social media research. Through large-scale data analysis, the study reveals new characteristics of public discourse in the digital age, challenges traditional agenda-setting theory, and provides new perspectives for understanding polarization phenomena on social media. Despite certain limitations, its methodological innovations and findings have important significance for advancing theoretical development and practical applications in related fields.