Conflicting Rationales, Shifting Stances: Unpacking User Divergence in Online Geopolitical Debates

Paper Title

Conflicting Rationales, Shifting Stances: Unpacking User Divergence in Online Geopolitical Debates

Publication Info

• Topic area: Online political discourse and user behavior in geopolitical conflict discussions.
• Keywords: Geopolitical conflicts, Russia-Ukraine war, Israel-Palestine war, social media, stance detection, rationale divergence, abusive content, echo chambers, content moderation, user consistency.

Background and Problem

• Problem / challenge: Existing research focuses on stance detection in online political discussions but overlooks the rationales users employ to justify their stances. This gap limits the effectiveness of moderation tools and dialogue-support systems.
• Significance: Understanding user rationales and stance divergence is critical for designing moderation tools, recommender systems, and interfaces that foster healthier online discourse and mitigate polarization.
• Motivation and related work: Prior studies have analyzed public opinion and polarization in individual conflicts, such as Russia-Ukraine and Israel-Palestine, but lack comparative analyses of rationales across conflicts. This study addresses the gap by examining user behavior across both conflicts simultaneously.

Solution

• Proposed approach: A rationale-aware, cross-conflict measurement framework using Reddit discussions to analyze user stances and rationales across the Russia-Ukraine and Israel-Palestine conflicts.
• Novelty:
  1. Development of a reusable annotation methodology for stance and rationale detection.
  2. Creation of a taxonomy of 11 rationales to classify user comments.
  3. Analysis of user divergence and consistency in rationales across two conflicts.
  4. Exploration of the relationship between echo chambers, abusive content, and user behavior.
• Procedure and key techniques:
  • Data collection: Extracted 50 million Reddit comments from Feb 2022–Dec 2024 using high-precision keywords for each conflict.
  • Stance detection: Aspect-Based Sentiment Analysis (ABSA) model to categorize comments as pro-Russia, pro-Ukraine, pro-Israel, pro-Palestine, or other stances.
  • Rationale detection: Zero-shot classification using a pre-trained language model to assign one of 11 rationales to each comment.
  • Analysis: Quantified stance combinations, rationale distributions, abusive content, echo chambers, and user consistency using statistical and network analysis.

Results

• Concrete findings:
  • Pro-Ukraine stances dominate the Russia-Ukraine conflict (44.81% of comments), while the Israel-Palestine conflict exhibits a more balanced stance distribution.
  • The most common rationale in the Russia-Ukraine conflict is "Geopolitical Strategy" (36.8%), while "Xenophobia and racism" dominates the Israel-Palestine conflict (28.4%).
  • Abusive content is higher in Russia-Ukraine discussions (5.29%) compared to Israel-Palestine (4.45%), with "Xenophobia and racism" being the most abusive rationale.
  • Echo chambers are more pronounced in Russia-Ukraine discussions, with lower entropy correlating with reduced abusive content.
  • Consistent users across conflicts exhibit higher proportions of abusive comments and tend to focus on identity-based rationales like "Xenophobia and racism" and "Cultural/Historical/National Identity."
• Advantage over baselines:
  • Introduced rationale-level annotation and analysis, which is absent in prior stance detection studies.
  • Demonstrated nuanced insights into user behavior across multiple conflicts, revealing patterns of consistency and divergence.
• Experiments / evaluation:
  • Validated stance and rationale detection models with manual annotations (accuracy: 0.77, Fleiss’ Kappa: 0.836 for stance detection, 0.853 for rationale detection).
  • Conducted network analysis to measure echo chamber effects and their correlation with abusive content.
• Limitations and future work:
  • Platform bias: Focused on English-speaking Reddit users, limiting global generalizability.
  • Methodological constraints: Single-label rationale classification and reliance on models trained on non-political domains.
  • Temporal dynamics: Aggregated data over two years, potentially obscuring short-term shifts in user behavior.

Summary

This study provides a comprehensive analysis of user stances and rationales in online discussions of the Russia-Ukraine and Israel-Palestine conflicts. It introduces a novel methodology for rationale-level annotation and cross-conflict comparison, revealing patterns of stance divergence, rationale usage, and abusive behavior. Findings highlight the role of echo chambers in mitigating abuse and the correlation between user consistency and toxicity. These insights inform the design of moderation tools and community strategies to balance safety, free expression, and diverse discourse in online environments. Future work should expand to multilingual platforms and explore hybrid rationale classification.