22 August 2022 in IEEE Transactions on Automatic Control
This article investigates a two-timescale opinion dynamics model, named the concatenated Friedkin–Johnsen (FJ) model, which describes the evolution of the opinions of a group of agents over a sequence of discussion events. The topology of the underlying graph changes with the event, in the sense that the agents can participate or less to an event, and the agents are stubborn, with stubbornness that can vary from one event to the other. Concatenation refers to the fact that the final opinions of an event become initial conditions of the next event. We show that a concatenated FJ model can be represented as a time-varying product of stochastic transition matrices having a special form. Conditions are investigated under which a concatenated FJ model can achieve consensus in spite of the stubbornness. Four different sufficient conditions are obtained, mainly based on the special topological structure of our stochastic matrices.
19 October 2021 in IEEE Transactions on Automatic Control
A biased assimilation model of opinion dynamics is a nonlinear model, in which opinions exchanged in a social network are multiplied by a state-dependent term having the bias as exponent and expressing the bias of the agents toward their own opinions. The aim of this article is to extend the bias assimilation model to signed social networks. We show that while for structurally balanced networks, polarization to an extreme value of the opinion domain (the unit hypercube) always occurs regardless of the value of the bias, for structurally unbalanced networks, a stable state of indecision (corresponding to the centroid of the opinion domain) also appears, at least for small values of the bias. When the bias grows and passes a critical threshold, which depends on the amount of “disorder” encoded in the signed graph, then a bifurcation occurs and opinions become again polarized.
29 November 2022 in IEEE Transactions on Automatic Control
Laplacian dynamics on signed digraphs have a richer behavior than those on nonnegative digraphs. In particular, for the so-called “repelling” signed Laplacians, the marginal stability property (needed to achieve consensus) is not guaranteed a priori and, even when it holds, it does not automatically lead to consensus, as these signed Laplacians may lose rank even in strongly connected digraphs. Furthermore, in the time-varying case, instability can occur even when switching in a family of systems each of which corresponds to a marginally stable signed Laplacian with the correct corank. In this paper we present novel conditions for achieving consensus on signed digraphs based on the property of eventual positivity, a Perron-Frobenius type of property for signed matrices. The conditions we develop cover both time-invariant and time-varying cases. A particularly simple sufficient condition, valid in both cases, is that the Laplacians are normal matrices. Such condition can be relaxed in several ways. For instance, in the time-invariant case it is enough that the Laplacian has this Perron-Frobenius property on the right side, but not on the left side (i.e., on the transpose). For the time-varying case, convergence to consensus can be guaranteed by the existence of a common Lyapunov function for all the signed Laplacians. All conditions can be easily extended to bipartite consensus.
1 November 2023 in PNAS Nexus
We consider complex multistage multiagent negotiation processes such as those occurring at climate conferences and ask ourselves how can an agent maximize its social power, intended as influence over the outcome of the negotiation. This question can be framed as a strategic game played over an opinion dynamics model, in which the action of an agent consists in stubbornly defending its own opinion. We show that for consensus-seeking opinion dynamics models in which the interaction weights are uniform, the optimal action obeys to an early mover advantage principle, i.e. the agents behaving stubbornly in the early phases of the negotiations achieve the highest social power. When looking at data collected from the climate change negotiations going on at the United Nations Framework Convention on Climate Change, we find evidence of the use of the early mover strategy. Furthermore, we show that the social powers computed through our model correlate very well with the perceived leadership roles assessed through independent survey data, especially when non-uniform weights incorporating economical and demographic factors are considered.
22 April 2024 in IEEE Transactions on Automatic Control
In this article, we propose and solve a social power game, i.e., a strategic game formulated on an opinion dynamics model and in which the agents aim to maximize their social power. As model we consider the concatenated Friedkin–Johnsen (FJ) model, which describes opinion evolution over a sequence of discussion events, while as actions we take the stubbornness coefficients, which can be freely chosen by the agents in order to maximize their social power, here corresponding to the utility function of the game. We show that the optimal solution of the social power game corresponds to an “early mover” strategy, in the sense that being stubborn as much as possible in early meetings allows to achieve the highest social power. This early mover advantage can be explained in terms of a diminishing return law that exists in the concatenated FJ model.
11 November 2025 in IEEE Transactions on Automatic Control
In this article, we consider a strategic game played by a group of agents on a set of opinion dynamics models. The models are all Friedkin–Johnsen (FJ) models, which are independent of each other (we call them “parallel FJ models”). The task of an agent is to maximize her overall social power by allocating a given budget of stubbornness across the parallel FJ models. For this game, the cost function is shown to be convex in the action profile set, but discontinuous at some boundary points when for some FJ model only one agent is stubborn (i.e., assigning nonzero stubbornness in the FJ model). Despite the discontinuity, an Nash equilibrium is shown to exist, but is not necessarily unique. Some sufficient conditions that can guarantee the uniqueness are proposed, relying on the strictly monotone pseudogradient mappings associated to the game. The conditions are applied to complete graphs with rank-1 weight matrices, for which the link weights are unequal for different agents and on different FJ models. Moreover, for the complete graph case, given the actions of the other agents, the best response of each agent is analytically characterized.