Modelling opinion dynamics in the age of algorithmic personalisation

Negative online news articles are shared more to social media

Prior research demonstrates that news-related social media posts using negative language are re-posted more, rewarding users who produce negative content. We investigate whether negative material from external news sites is also introduced to social media through more user posts, offering comparable incentives for journalists to adopt a negative tone. Data from four US and UK news sites (95,282 articles) and two social media platforms (579,182,075 posts on Facebook and Twitter, now X) show social media users are 1.91 times more likely to share links to negative news articles. The impact of negativity varies by news site and social media platform and, for political articles, is moderated by topic focus, with users showing a greater inclination to share negative articles referring to opposing political groups. Additionally, negativity amplifies news dissemination on social media to a greater extent when accounting for the re-sharing of user posts containing article links. These findings suggest a higher prevalence of negatively toned articles on Facebook and Twitter compared to online news sites. Further, should journalists respond to the incentives created by the heightened sharing of negative articles to social media platforms, this could even increase negative news exposure for those who do not use social media.

Followers do not dictate the virality of news outlets on social media

Initially conceived for entertainment, social media platforms have profoundly transformed the dissemination of information and consequently reshaped the dynamics of agenda-setting. In this scenario, understanding the factors that capture audience attention and drive viral content is crucial. Employing Gibrat's Law, which posits that an entity's growth rate is unrelated to its size, we examine the engagement growth dynamics of news outlets on social media. Our analysis includes the Facebook historical data of over a thousand news outlets, encompassing approximately 57 million posts in four European languages from 2008 to the end of 2022. We discover universal growth dynamics according to which news virality is independent of the traditional size of the outlet. Moreover, our analysis reveals a significant long-term impact of news source reliability on engagement growth, with engagement induced by unreliable sources decreasing over time. We conclude the article by presenting a statistical model replicating the observed growth dynamics.

Bursts of communication increase opinion diversity in the temporal Deffuant model

Human interactions create social networks forming the backbone of societies. Individuals adjust their opinions by exchanging information through social interactions. Two recurrent questions are whether social structures promote opinion polarisation or consensus and whether polarisation can be avoided, particularly on social media. In this paper, we hypothesise that not only network structure but also the timings of social interactions regulate the emergence of opinion clusters. We devise a temporal version of the Deffuant opinion model where pairwise social interactions follow temporal patterns. Individuals may self-organise into a multi-partisan society due to network clustering promoting the reinforcement of local opinions. Burstiness has a similar effect and is alone sufficient to refrain the population from consensus and polarisation by also promoting the reinforcement of local opinions. The diversity of opinions in socially clustered networks thus increases with burstiness, particularly, and counter-intuitively, when individuals have low tolerance and prefer to adjust to similar peers. The emergent opinion landscape is well-balanced regarding groups' size, with relatively short differences between groups, and a small fraction of extremists. We argue that polarisation is more likely to emerge in social media than offline social networks because of the relatively low social clustering observed online, despite the observed online burstiness being sufficient to promote more diversity than would be expected offline. Increasing the variance of burst activation times, e.g. by being less active on social media, could be a venue to reduce polarisation. Furthermore, strengthening online social networks by increasing social redundancy, i.e. triangles, may also promote diversity.

Effect of collaborative-filtering-based recommendation algorithms on opinion polarization

A central role in shaping the experience of users online is played by recommendation algorithms. On the one hand they help retrieving content that best suits users taste, but on the other hand they may give rise to the so-called "filter bubble" effect, favoring the rise of polarization. In the present paper we study how a user-user collaborative-filtering algorithm affects the behavior of a group of agents repeatedly exposed to it. By means of analytical and numerical techniques we show how the system stationary state depends on the strength of the similarity and popularity biases, quantifying respectively the weight given to the most similar users and to the best rated items. In particular, we derive a phase diagram of the model, where we observe three distinct phases: disorder, consensus, and polarization. In the last users spontaneously split into different groups, each focused on a single item. We identify, at the boundary between the disorder and polarization phases, a region where recommendations are nontrivially personalized without leading to filter bubbles. Finally, we show that our model well reproduces the behavior of users on the online music platform last.fm. This analysis paves the way to a systematic analysis of recommendation algorithms by means of statistical physics methods and opens the possibility of devising less polarizing recommendation algorithms.

Maximum entropy networks for large scale social network node analysis

Recently proposed computational techniques allow the application of various maximum entropy network models at a larger scale. We focus on disinformation campaigns and apply different maximum entropy network models on the collection of datasets from the Twitter information operations report. For each dataset, we obtain additional Twitter data required to build an interaction network. We consider different interaction networks which we compare to an appropriate null model. The null model is used to identify statistically significant interactions. We validate our method and evaluate to what extent it is suited to identify communities of members of a disinformation campaign in a non-supervised way. We find that this method is suitable for larger social networks and allows to identify statistically significant interactions between users. Extracting the statistically significant interaction leads to the prevalence of users involved in a disinformation campaign being higher. We found that the use of different network models can provide different perceptions of the data and can lead to the identification of different meaningful patterns. We also test the robustness of the methods to illustrate the impact of missing data. Here we observe that sampling the correct data is of great importance to reconstruct an entire disinformation operation.

Opinion Polarization in Human Communities Can Emerge as a Natural Consequence of Beliefs Being Interrelated

The emergence of opinion polarization within human communities-the phenomenon that individuals within a society tend to develop conflicting attitudes related to the greatest diversity of topics-has been a focus of interest for decades, both from theoretical and modelling points of view. Regarding modelling attempts, an entire scientific field-opinion dynamics-has emerged in order to study this and related phenomena. Within this framework, agents' opinions are usually represented by a scalar value which undergoes modification due to interaction with other agents. Under certain conditions, these models are able to reproduce polarization-a state increasingly familiar to our everyday experience. In the present paper, an alternative explanation is suggested along with its corresponding model. More specifically, we demonstrate that by incorporating the following two well-known human characteristics into the representation of agents: (1) in the human brain beliefs are interconnected, and (2) people strive to maintain a coherent belief system; polarization immediately occurs under exposure to news and information. Furthermore, the model accounts for the proliferation of fake news, and shows how opinion polarization is related to various cognitive biases.

FOMO (fate of online media only) in infectious disease modeling: a review of compartmental models

Mathematical models played in a major role in guiding policy decisions during the COVID-19 pandemic. These models while focusing on the spread and containment of the disease, largely ignored the impact of media on the disease transmission. Media plays a major role in shaping opinions, attitudes and perspectives and as the number of people online increases, online media are fast becoming a major source for news and health related information and advice. Consequently, they may influence behavior and in due course disease dynamics. Unlike traditional media, online media are themselves driven and influenced by their users and thus have unique features. The main techniques used to incorporate online media mathematically into compartmental models, with particular reference to the ongoing COVID-19 pandemic are reviewed. In doing so, features specific to online media that have yet to be fully integrated into compartmental models such as misinformation, different time scales with regards to disease transmission and information, time delays, information super spreaders, the predatory nature of online media and other factors are identified together with recommendations for their incorporation.

Depolarization of echo chambers by random dynamical nudge

In social networks, users often engage with like-minded peers. This selective exposure to opinions might result in echo chambers, i.e., political fragmentation and social polarization of user interactions. When echo chambers form, opinions have a bimodal distribution with two peaks on opposite sides. In certain issues, where either extreme positions contain a degree of misinformation, neutral consensus is preferable for promoting discourse. In this paper, we use an opinion dynamics model that naturally forms echo chambers in order to find a feedback mechanism that bridges these communities and leads to a neutral consensus. We introduce the random dynamical nudge (RDN), which presents each agent with input from a random selection of other agents’ opinions and does not require surveillance of every person’s opinions. Our computational results in two different models suggest that the RDN leads to a unimodal distribution of opinions centered around the neutral consensus. Furthermore, the RDN is effective both for preventing the formation of echo chambers and also for depolarizing existing echo chambers. Due to the simple and robust nature of the RDN, social media networks might be able to implement a version of this self-feedback mechanism, when appropriate, to prevent the segregation of online communities on complex social issues.

A general framework to link theory and empirics in opinion formation models

We introduce a minimal opinion formation model that is quite flexible and can reproduce a wide variety of the existing micro-influence assumptions and models. The model can be easily calibrated on real data, upon which it imposes only a few requirements. From this perspective, our model can be considered as a bridge, connecting theoretical studies on opinion formation models and empirical research on social dynamics. We investigate the model analytically by using mean-field approximation and numerically via Monte Carlo simulations. Our analysis is exemplified by recently reported empirical data drawn from an online social network. We demonstrate that the model calibrated on these data may reproduce fragmented and polarizing social systems. Furthermore, we manage to generate an artificial society that features properties quantitatively and qualitatively similar to those observed empirically at the macro scale. This ability became possible after we had advanced the model with two important communication features: selectivity and personalization algorithms.

Filter bubble effect in the multistate voter model

Social media influence online activity by recommending to users content strongly correlated with what they have preferred in the past. In this way, they constrain users within filter bubbles strongly limiting their exposure to new or alternative content. We investigate this type of dynamics by considering a multistate voter model where, with a given probability λ, a user interacts with "personalized information," suggesting the opinion most frequently held in the past. By means of theoretical arguments and numerical simulations, we show the existence of a nontrivial transition between a region (for small λ) where a consensus is reached and a region (above a threshold λ_c) where the system gets polarized and clusters of users with different opinions persist indefinitely. The threshold always vanishes for large system size N, showing that a consensus becomes impossible for a large number of users. This finding opens new questions about the side effects of the widespread use of personalized recommendation algorithms.

Algorithmic amplification of politics on Twitter

Content on Twitter's home timeline is selected and ordered by personalization algorithms. By consistently ranking certain content higher, these algorithms may amplify some messages while reducing the visibility of others. There's been intense public and scholarly debate about the possibility that some political groups benefit more from algorithmic amplification than others. We provide quantitative evidence from a long-running, massive-scale randomized experiment on the Twitter platform that committed a randomized control group including nearly 2 million daily active accounts to a reverse-chronological content feed free of algorithmic personalization. We present two sets of findings. First, we studied tweets by elected legislators from major political parties in seven countries. Our results reveal a remarkably consistent trend: In six out of seven countries studied, the mainstream political right enjoys higher algorithmic amplification than the mainstream political left. Consistent with this overall trend, our second set of findings studying the US media landscape revealed that algorithmic amplification favors right-leaning news sources. We further looked at whether algorithms amplify far-left and far-right political groups more than moderate ones; contrary to prevailing public belief, we did not find evidence to support this hypothesis. We hope our findings will contribute to an evidence-based debate on the role personalization algorithms play in shaping political content consumption.

Effect of algorithmic bias and network structure on coexistence, consensus, and polarization of opinions

Individuals of modern societies share ideas and participate in collective processes within a pervasive, variable, and mostly hidden ecosystem of content filtering technologies that determine what information we see online. Despite the impact of these algorithms on daily life and society, little is known about their effect on information transfer and opinion formation. It is thus unclear to what extent algorithmic bias has a harmful influence on collective decision-making, such as a tendency to polarize debate. Here we introduce a general theoretical framework to systematically link models of opinion dynamics, social network structure, and content filtering. We showcase the flexibility of our framework by exploring a family of binary-state opinion dynamics models where information exchange lies in a spectrum from pairwise to group interactions. All models show an opinion polarization regime driven by algorithmic bias and modular network structure. The role of content filtering is, however, surprisingly nuanced; for pairwise interactions it leads to polarization, while for group interactions it promotes coexistence of opinions. This allows us to pinpoint which social interactions are robust against algorithmic bias, and which ones are susceptible to bias-enhanced opinion polarization. Our framework gives theoretical ground for the development of heuristics to tackle harmful effects of online bias, such as information bottlenecks, echo chambers, and opinion radicalization.

Characterizing network dynamics of online hate communities around the COVID-19 pandemic

Hate speech has long posed a serious problem for the integrity of digital platforms. Although significant progress has been made in identifying hate speech in its various forms, prevailing computational approaches have tended to consider it in isolation from the community-based contexts in which it spreads. In this paper, we propose a dynamic network framework to characterize hate communities, focusing on Twitter conversations related to COVID-19 in the United States and the Philippines. While average hate scores remain fairly consistent over time, hate communities grow increasingly organized in March, then slowly disperse in the succeeding months. This pattern is robust to fluctuations in the number of network clusters and average cluster size. Infodemiological analysis demonstrates that in both countries, the spread of hate speech around COVID-19 features similar reproduction rates as other COVID-19 information on Twitter, with spikes in hate speech generation at time points with highest community-level organization of hate speech. Identity analysis further reveals that hate in the US initially targets political figures, then grows predominantly racially charged; in the Philippines, targets of hate consistently remain political over time. Finally, we demonstrate that higher levels of community hate are consistently associated with smaller, more isolated, and highly hierarchical network clusters across both contexts. This suggests potentially shared structural conditions for the effective spread of hate speech in online communities even when functionally targeting distinct identity groups. Our findings bear theoretical and methodological implications for the scientific study of hate speech and understanding the pandemic's broader societal impacts both online and offline.

The ethics of algorithms: key problems and solutions

Research on the ethics of algorithms has grown substantially over the past decade. Alongside the exponential development and application of machine learning algorithms, new ethical problems and solutions relating to their ubiquitous use in society have been proposed. This article builds on a review of the ethics of algorithms published in 2016 (Mittelstadt et al. Big Data Soc 3(2), 2016). The goals are to contribute to the debate on the identification and analysis of the ethical implications of algorithms, to provide an updated analysis of epistemic and normative concerns, and to offer actionable guidance for the governance of the design, development and deployment of algorithms.

A network model for polarization of political opinion

We propose and study a simple model for the evolution of political opinion through a population. The model includes a nonlinear term that causes individuals with more extreme views to be less receptive to external influence. Such a term was suggested in 1981 by Cobb in the context of a scalar-valued diffusion equation, and recent empirical studies support this modeling assumption. Here, we use the same philosophy in a network-based model. This allows us to incorporate the pattern of pairwise social interactions present in the population. We show that the model can admit two distinct stable steady states. This bi-stability property is seen to support polarization and can also make the long-term behavior of the system extremely sensitive to the initial conditions and to the precise connectivity structure. Computational results are given to illustrate these effects.