Evidence suggests that consumers prefer media products that align with their political beliefs. Hence, a media organization is economically incentivized to tailor their output to the political leanings of their consumers. At the same time, consumers shape their perceptions of the political bias of a media organization by reacting to published outputs (e.g. daily newspaper editorials) and observing each other's opinions (e.g. as conveyed on social media). Hence, there is a two-way feedback loop between networked consumers and the media organization, which controls how media bias and people's perceptions about it co-evolve.

To model the above scenario, the complex problem of inferring the political bias of a media organization is mapped onto the idealized problem of inferring the evolving bias of a coin. A network of Bayesian learners is constructed, where the beliefs of each agent about the coin bias obey a probability distribution. The beliefs are updated in response to (i) observations of the coin toss, and (ii) “peer pressure” from political allies (positive ties) and opponents (negative ties). Simultaneously, the coin (media organization) “surveys” the agents by sampling their beliefs and shifts its bias towards the sample mean. Monte Carlo multi-agent simulations are conducted to study the co-evolution of the bias and beliefs and answer important two-sided questions about the long-term behavior, e.g., (i) does the bias stabilize or fluctuate? And (ii) do the beliefs exhibit turbulent nonconvergence or intermittency, as for a static coin?

Modelling signed polytomous network data is crucial for understanding complex relational structures in systems where edges between nodes vary not only in strength (multiple weight categories) but also in nature (positive or negative associations). Such networks are common in fields like social networks, political alliances, and collaborative systems, where interactions can have both supportive and adversarial dynamics.

We introduce a novel framework based on exponential-family random graph models for signed polytomous networks, where the joint probability of the network is modelled as the product of two components: a baseline categorical process that captures the unweighted signed connectivity structure, and a conditional dissolution process that models the dynamics of edge dissolution based on the polarity of connections. To estimate model parameters and make inferences, we adopt a comprehensive Bayesian approach, offering flexibility in modelling and robust uncertainty quantification. We demonstrate the effectiveness of our model through empirical analysis on real-world signed polytomous networks, showcasing its utility for relational inferential tasks.

We introduce a model of information dissemination in signed networks. It is a discrete-time process in which uninformed actors incrementally receive information from their informed neighbors or from the outside. Our goal is to minimize the number of confused actors - that is, the number of actors who receive contradictory information.

We prove upper bounds for the number of confused actors in signed networks and in equivalence classes of signed networks.

In particular, we show that there are signed networks where, for any information placement strategy, almost 60\% of the actors are confused. Furthermore, this is also the case when considering the minimum number of confused actors within an equivalence class of signed graphs

Echo chamber effects in social networks are generally attributed to the prevalence of interactions among like-minded peers. However, recent evidence has emphasized the role of hostile interactions between opposite-minded groups. We investigate the role of polarization, identified with structural balance, in the formation of echo chambers in signed networks. To do so, we generalize the Independent Cascade Model and the Linear Threshold Model to describe information propagation in presence of negative edges. Antagonistic connections do not disrupt the flow of information, but instead, alter the way information is framed. Our results show that echo chambers spontaneously emerge in balanced networks, but also in antibalanced ones for specific parameters. This highlights that structural polarization and echo chambers do not necessarily display a one-to-one correspondence, showing instead a complex and often counterintuitive interplay. The robustness of our results is confirmed with a complex contagion model and through simulations in different network topologies, including real-world datasets.

The role of social ties on health and well-being is well-documented, but the association between antagonistic ties and well-being is incompletely understood. Using longitudinal data of health outcomes and sociocentrically mapped networks of 19,777 residents of 176 villages in the rural highlands of Honduras, we explore how the existence of, and spatial proximity to, antagonistic social ties is associated with subsequent mental and physical health outcomes. We find that while the existence of incoming antagonistic ties does not affect mental or physical health, both the existence of outgoing negative ties, and their proximity to respondents, is associated with self-reported mental health two years later. Social interactions can be not only salubrious, but also deleterious, which can have far-reaching implications for well-being.