Exploring the Role of Randomization on Belief Rigidity in Online Social Networks

TL;DR

This study investigates how randomized exposure in online social networks affects belief rigidity around climate-change policy prompts. Using a two-condition experimental framework (similar-peer vs randomized recommendations) with five rounds and 163 participants, the authors quantify private belief changes via a Delphi-like revision stage and measure peer influence through social signals, follow relationships, and semantic similarity of reasoning. They find that peer opinions influence beliefs under both conditions, belief updates occur in roughly 25% of cases with stronger coupling when changes happen, and that randomized recommendations modestly increase the uptake of diverse viewpoints while reducing homophily. The findings offer practical insights for platform design to combat echo chambers and polarization and establish a versatile experimental framework for testing future interventions targeting belief rigidity and related cues.

Abstract

People often stick to their existing beliefs, ignoring contradicting evidence or only interacting with those who reinforce their views. Social media platforms often facilitate such tendencies of homophily and echo-chambers as they promote highly personalized content to maximize user engagement. However, increased belief rigidity can negatively affect real-world policy decisions such as leading to climate change inaction and increased vaccine hesitancy. To understand and effectively tackle belief rigidity on online social networks, designing and evaluating various intervention strategies is crucial, and increasing randomization in the network can be considered one such intervention. In this paper, we empirically quantify the effects of a randomized social network structure on belief rigidity, specifically examining the potential benefits of introducing randomness into the network. We show that individuals' beliefs are positively influenced by peer opinions, regardless of whether those opinions are similar to or differ from their own by passively sensing belief rigidity through our experimental framework. Moreover, people incorporate a slightly higher variety of different peers (based on their opinions) into their networks when the recommendation algorithm provides them with diverse content, compared to when it provides them with similar content. Our results indicate that in some cases, there might be benefits to randomization, providing empirical evidence that a more randomized network could be a feasible way of helping people get out of their echo-chambers. Our findings have broader implications in computing and platform design of social media, and can help combat overly rigid beliefs in online social networks.

Figures (7)

• Figure 1: In this figure, we describe the experimental framework, along with examples to understand the flow of the experiment. Condition 1 is designed to show similar users, while condition 2 is designed to show random users. Participants start by agreeing or disagreeing to a statement on a 7 point Likert Scale in stage 1. We denote a participant rating agreeably to the statement as green and rating disagreeably as red, while orange is neutral. In stage 2, for condition 1, when a participant rates the statement as green, they mostly see others who have rated it green or orange. A similar scenario would occur if they participant rated the statement red, as they would mostly see other reds or oranges. In condition 2, however, participants see a wide array of ratings. An example is provided in this figure, where even though the participant disagreed (denoted by red), they see users with red, orange and green ratings. Stage 3 is designed for network restructuring as participants can follow (denoted by green lines) and unfollow (denoted by red lines) people or stick to their previous choices (denoted by black lines).
• Figure 2: (a) Percentage of times participants changed Likert ratings in Condition 1 (similar condition). Participants changed Likert ratings 22.86 percent compared to not changing 77.14 percent of the time. (b) Percentage of times participants changed Likert ratings in Condition 2 (randomized condition). Participants changed Likert ratings 27.36 percent time compared to not changing 72.64 percent of the time.
• Figure 3: (a) Histogram of mean change in Likert rating for each individual in Condition 1 (similar condition) (b) Histogram of mean change in Likert rating for each individual in Condition 2 (randomized condition) In both cases, our results show that individuals are mostly rigid in their beliefs and do not drastically change them.
• Figure 4: (a) Distribution of mean change in Likert rating in Condition 1 (similar condition) (b) Distribution of mean change in Likert rating in Condition 2 (randomized condition) Both distributions peak at 0 and are unimodal.
• Figure 5: (a) Belief Network Distance for Condition 1 (similar condition). Mean Followed (denoted by blue line) is not statistically different from mean Not followed (denoted by orange line). (b) Belief Network Distance for Condition 2 (randomized condition). Mean Followed (denoted by blue line) is lower than mean Not followed (denoted by orange line), signifying people's homophilic tendencies. However, mean Followed in C2 is greater than mean Followed in C1, showing people's tendency to incorporate a wider variety of beliefs to some extent.