Social networks 2

As people increasingly turn to online platforms for learning and information exchange, understanding how information is transformed through repeated transmission is essential for predicting collective outcomes and mitigating misinformation. Research on social learning and collective cognition highlights that the validity of information received by individuals is strongly influenced by the pathways through which it travels. While stories passed through many individuals often become distorted, information cross-checked against multiple references at various stages tends to be more accurate.In a novel, large-scale preregistered behavioral experiment, we demonstrate that the structural features of social networks significantly impact the fidelity and divisiveness of transmitted information. Participants (n=864) read and reproduced a seed text about the history and risks of antibiotics under two experimental conditions: a single-pathway network, where information flowed linearly, and a multiple-pathway network, where information was redundantly transmitted through three pathways (Fig. 1A–B).Our findings show that even minimal structural redundancy can enhance information fidelity and promote consensus in the retelling of a scientific text. Specifically, transmission in multiple-pathway networks (i) reduces information loss as texts propagate between participants (Fig. 1C), and (ii) promotes consensus on information content among non-interacting participants in independently tested networks (Fig. 1D).To further understand these patterns, we developed a simple computational model grounded in prior research on social learning. This model provides a mechanistic explanation of how redundancy influences information transmission. Simulations using the model qualitatively replicate the experimental results (Fig. 1D). Finally, exploratory analyses reveal that redundancy amplifies the transmission of emotional content, particularly fear-related information.

We present a Python library for simulating an opinion model that abstracts the communication dynamics within a social media platform called DOCES (Dynamical Opinion Clusters Exploration Suite). Its core functions are implemented in CPython to achieve C performance. In this model, agents have continuous opinions with values between -1 and 1. The agents can share external content in the social network, simulating real-world observations. In addition, they can re-post information based on their feed. A social network's recommendation algorithm governs the reception of content by peers. Agents then adjust their opinions based on the content they receive. Disagreements in opinions can lead to changes in friendships. Each step is governed by probabilistic filters that take user opinions into account. In addition to its configurations, DOCES allows the measurement of several outputs of the dynamics, including the resulting network structure and the opinions of the users. We used DOCES to study the effects of changing the verification policy on X, which is called X Premium. We set up DOCES to mimic the behavior of the algorithm for verified users, henceforth called priority users. We adapted the model so that priority users always have their posts delivered to their followers. We found that the echo chambers weaken by increasing the number of priority users. We also tested the possibility of extremists taking advantage of verified accounts. We named them as ideologues. In this case, we set half to opinion -1 and the other half to 1. We found that the priority accounts can significantly amplify the echo chambers. We also varied the percentage of priority users and ideologues. As expected, the echo chambers tend to weaken as the number of priority users increases. Surprisingly, though, the echo chambers are amplified after sufficient ideologues are added, showing that a few percent can make a difference.

Political polarization has become an increasingly concerning societal phenomenon in recent years, with far-reachingimplications for democratic discourse, policy making and social cohesion. As traditional public squares have given wayto digital arenas, social media platforms have emerged as primary venues for political debate and opinion formation.However, the impact of these digital spaces on polarization dynamics remains a subject of ongoing research and debate.This raises fundamental questions about the nature of online political discourse: Are all digital spaces where politicaldiscussion takes place destined to become hierarchical, politically divided, and develop echo chambers?While much attention has focused on major social media platforms, new digital spaces are emerging that not onlyshow potential for fostering less toxic environments but are also built on decentralized architecture. This study contributesto the growing body of knowledge on social media polarization by examining Bluesky, a relatively new platform. Ourresearch measures both structural [1] and affective [2] polarization within the Bluesky ecosystem. We analyze three keyaspects of each selected political communication network, which represents a prominent topic of discussion within theplatform: group structure through stochastic block models, issue-based polarization with the adaptive EI-index, and issuealignment using normalized mutual information. Through this analysis, we address the following questions: 1) To whatextent does polarization exist on Bluesky, and what distinct polarized groups emerge? 2) What are the dominant topicsof discussion on Bluesky, and which topics show the highest levels of polarization? 3) How do these patterns compare topreviously observed polarization on similar platforms?The full abstract is in the attached PDF.

The entertainment-driven dynamics of social media platforms encourage users to engage with like-minded individuals and consume content aligned with their beliefs.These dynamics may amplify polarization by reinforcing shared perspectives and reducing exposure to diverse viewpoints.Simultaneously, users migrate from one platform to another, either forced by moderation policies, such as de-platforming, or spontaneously seeking environments more aligned with their preferences. These migrations foster the specialization and differentiation of the social media ecosystem, with platforms increasingly organized around specific user communities and shared content preferences. This shift marks an evolution from echo chambers enclosed within platforms to "echo platforms", i.e., entire platforms functioning as ideologically homogeneous niches. This study introduces an operational framework to systematically analyze these dynamics, by examining three key dimensions: platform centrality (central vs. peripheral), news consumption (reliable vs questionable), and user base composition (uniform vs diverse). To this aim, we leverage a dataset of 126M URLs posted by nearly 6M users on nine social media platforms, namely Facebook, Reddit, Twitter (now X), YouTube, BitChute, Gab, Parler, Scored, and Voat. We model traffic between the nine platforms, and between platform and external domains, using weighted directed networks, and utilize apt null models to account for platforms' size disparity when needed. We find a clear separation between mainstream and alt-tech platforms, with the second category being characterized by a peripheral role in the social media ecosystem, a greater prevalence of unreliable content, and a heightened ideological uniformity. These findings outline the main dimensions defining the fragmentation and polarization of the social media ecosystem.

A tenet of the last century of social science is that the wiring of social networks explains their function. It has been argued that people are not merely passively affected by their positions in social networks—instead, we try to exploit and improve our network positions. Artificially intelligent agents entering a world of humans—either as commercial applications viaexperiments in the social and behavioral sciences—need a sense of how humans operate theirsocial networks. But do they have it yet? In this work, we challenge LLMs and humans to imagine the social structures observed in three classical social network studies.Our findings reveal that humans show a marginally better understanding of social networks than the LLMs of 2024; at least in the third-person perspective that we investigate. On the other hand, even humans perform poorly, like the AI putting too much emphasis on external information (like corporate structure). Some interesting details are emerging, like how poorly both humans and machines predict the average degree and that humans don’t overestimate the managers asking their subordinates for advice or calling them friends.The difficulty of this task raises questions about AI's ability to replicate human social intuition. Could human sociality, deeply ingrained through evolution, provide an edge over machines? Maybe human sociality, and thus social network understanding, is hardcoded in our genes giving us an edge over the machines. Still, the LLMs have presumably been trained on a vast amount of literature about social networks, and could for that reason be expected to perform as well as people.The main question remaining, and an avenue for future research is if changing the perspective to the first person also changes the performance of both humans and machines. We have high hopes for the future of AIs as a tool to understand humans, but that will require us to know that AIs can predict and respond to human behavior close to perfectly.

Coalitions are social mechanisms for political actors to organise joint action for mutual gain. As coalitions drive collective action across political phenomena--including government formation, international relations, and public policy--understanding how they emerge and contribute to diverse outcomes, such as perceptions of political opponents, policy gridlocks, or even reshuffling of the global order is core to the study of politics. However, exactly how to conceptualise and operationalise coalitions has been challenging.Here, we introduce the complex coalition approach. It emphasises the notion of the relatively local social ties among political actors giving rise to political alliances. However, such emergent coalitions unfold not only within, but across various relational contexts, which provide actors with varying affordances to engage with each other and advance their goals. To accommodate these ideas under a single, statistically rigorous, and generalisable framework, we build on layered political networks and inferential network partitioning techniques to model the structure and dynamics of coalitions across spatially and temporally dependent relational contexts.To demonstrate the application and merits of our approach, we study coalition emergence in Finnish climate politics, focusing on formal collaboration, media discourse, and online endorsement among a hundred political organisations, both before and after the Paris Agreement. Specifically, we show how a 14-layer model that appropriately accounts for cross-context dependencies yields a substantially smaller description length (i.e., requires less information to describe the data) than a model that considers each of these relational contexts in isolation, suggesting that cross-context dependencies contribute substantial explanatory power to coalition emergence.

Existing studies of political polarization are often limited to a single country and one form of polarization, hindering a comprehensive understanding of the phenomenon. Here we investigate patterns of polarization online across nine countries (Canada, France, Germany, Italy, Poland, Spain, Turkey, UK, USA), focusing on the structure of political interaction networks, the use of toxic language targeting out-groups, and how these factors relate to user engagement. First, we show that political interaction networks are structurally polarized on Twitter (currently X; see Figure 1). Second, we reveal that out-group interactions, defined by the network, are more toxic than in-group interactions, indicative of affective polarization. Third, we show that out-group interactions receive lower engagement than in-group interactions. Finally, we identify a common ally-enemy structure in political interactions, show that political mentions are more toxic than apolitical mentions, and highlight that interactions between politically engaged accounts are limited and rarely reciprocated. These results hold across countries and represent a step towards a stronger cross-country understanding of polarization.