Balancing Opposing Views to Reduce Controversy

Kiran Garimella, Gianmarco De Francisci Morales, Aristides Gionis, Michael MathioudakisarXiv 2016 · arXiv

Abstract

Social media platforms often foster "echo chambers" where users are exposed primarily to views that reinforce their existing biases. This paper presents an algorithmic framework for reducing such polarization by recommending "bridge" connections between opposing sides. By introducing the Random Walk Controversy (RWC) metric and the Recommend Opposing View (ROV) algorithm, the authors demonstrate how targeting influential users can bridge divides. Furthermore, the ROV-AP variant incorporates acceptance probability, ensuring recommendations are not only mathematically optimal but also practically realistic for users.

Introduction

The "filter bubble" effect on social media has become a significant societal challenge, leading to increased hostility and isolation between groups with differing viewpoints. While much research has focused on content-based interventions, this work approaches the problem through the lens of network structure. The goal is to identify which new connections (edges) in an endorsement graph—where nodes represent users and edges represent retweets—would most effectively reduce the overall controversy of the network.

Visualization of endorsement graphs for different controversial topics, showing distinct clusters.

As shown in Figure 1, controversial discussions typically manifest as two disjoint clusters with very few cross-edges. The authors treat the reduction of this polarization as an edge-recommendation problem, balancing the potential impact of a connection with the likelihood that a user will accept the recommendation.

Methodology

Measuring Controversy: Random Walk Controversy (RWC)

To quantify polarization, the authors utilize the Random Walk Controversy (RWC) score. This metric measures the probability that a random walk starting in one community will stay within that community rather than crossing over to the other side. A high RWC score indicates a highly segregated and controversial network.

The ROV and ROV-AP Algorithms

The primary algorithmic contribution is ROV (Recommend Opposing View). This approach is based on the observation that polarized networks often resemble two "star graphs" with central influencers. ROV focuses on creating edges between the high-degree "centers" of these opposing sides.

Recognizing that users are unlikely to follow extreme opposites, the authors developed ROV-AP (ROV with Acceptance Probability). This version uses a machine-learning model to predict the probability of a user accepting a recommendation based on their "polarity score"—a measure of how extreme their views are. By optimizing for the expected reduction in RWC, ROV-AP identifies "bridgeable" users who are on opposing sides but share more common ground.

Computational Efficiency

A major hurdle in this approach is the $O(n^2)$ cost of recalculating the RWC score for every potential edge. The authors overcome this by applying the Sherman-Morrison formula, allowing for incremental updates to the RWC score. This technique resulted in speedups ranging from 5x to 65x compared to standard calculation methods.

Key Results

The algorithms were tested on ten real-world Twitter datasets, including topics like #guncontrol, #ukraine, and #obamacare.

Reduction in RWC score for the #ukraine dataset as edges are added by various algorithms.

As illustrated in Figure 3, the proposed methods (ROV and ROV-AP) significantly outperform structural baselines such as NetGel and Shortcut. While the "Greedy" approach often yields the highest mathematical reduction in controversy, it is computationally prohibitive for large networks without the optimizations introduced in this work.

Key findings include:

  • Influencer Impact: Connecting leaders of opposing sides is the most effective way to create a "trickle-down" bridging effect for their followers.
  • Acceptance Trade-off: ROV-AP successfully identifies connections that are more likely to be accepted than those suggested by the pure ROV model, often recommending users with higher content overlap.
  • Polarity as a Signal: A user's political polarity was found to be a highly effective predictor for link recommendation, rivaling more complex state-of-the-art algorithms.

Conclusion

This research provides a robust framework for proactively reducing social media controversy through structural interventions. By balancing the mathematical impact of new connections with the practical reality of user behavior, the ROV-AP algorithm offers a viable path toward bursting filter bubbles. Future work will explore multi-sided controversies and the dynamic nature of social networks to further refine these recommendation strategies.

Summary updated

BibTeX
@misc{garimella2016balancing,
  author = {Kiran Garimella and Gianmarco De Francisci Morales and Aristides Gionis and Michael Mathioudakis},
  title = {{Balancing Opposing Views to Reduce Controversy}},
  year = {2016},
  eprint = {1611.00172},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1611.00172},
}
Figure