Transitions between polarization and radicalization in a temporal bilayer echo-chamber model

Unanimity, Coexistence, and Rigidity: Three Sides of Polarization

Political polarization is perceived as a threat to democracies. Using the Galam model of opinion dynamics deployed in a five-dimensional parameter space, I show that polarization is the byproduct of an essential hallmark of a vibrant democratic society, namely open and informal discussions among agents. Indeed, within a homogeneous social community with floaters, the dynamics lead gradually toward unanimity (zero entropy). Polarization can eventually appear as the juxtaposition of non-mixing social groups sharing different prejudices about the issue at stake. On the other hand, the inclusion of contrarian agents produces a polarization within a community that mixes when their proportion x is beyond a critical value xc=16≈0.167 for discussing groups of size three and four. Similarly, the presence of stubborn agents also produces a polarization of a community that mixes when the proportion of stubborn agents is greater than some critical value. For equal proportions of stubborn agents a along each opinion, ac=29≈0.22 for group size four against ac=14=0.25 for group size three. However, the evaluation of the proportion of individual opinion shifts at the attractor 12 and indicates that the polarization produced by contrarians is fluid with a good deal of agents who keep shifting between the two opposed blocks (high entropy). That favors a coexistence of opposite opinions in a divided community. In contrast, the polarization created by stubborn agents is found to be frozen with very few individuals shifting opinion between the two opinions (low entropy). That yields a basis for the emergence of hate between the frozen opposed blocks.

Mean-field approximation for structural balance dynamics in heat bath

A critical temperature for a complete signed graph of N agents where the time-dependent polarization of links tends towards the Heider (structural) balance is found analytically using the heat-bath approach and the mean-field approximation as T^{c}=(N-2)/a^{c}, where a^{c}≈1.71649. The result is in perfect agreement with numerical simulations starting from the paradise state where all links are positively polarized as well as with the estimation of this temperature received earlier with much more sophisticated methods. When heating the system, one observes a discontinuous and irreversible phase transition at T^{c} from a nearly balanced state when the mean link polarization is about x_{c}=0.796388 to a disordered and unbalanced state where the polarization vanishes. When the initial conditions for the polarization of links are random, then at low temperatures a balanced bipolar state of two mutually hostile cliques exists that decays towards the disorder and there is a discontinuous phase transition at a temperature T^{d} that is lower than T^{c}. The system phase diagram corresponds to the so-called fold catastrophe when a stable solution of the mean-field equation collides with a separatrix, and as a result a hysteresislike loop is observed.

Topic selectivity and adaptivity promote spreading of short messages

Why is the Twitter, with its extremely length-limited messages so popular ? Our work shows that short messages focused on a single topic may have an inherent advantage in spreading through social networks, which may explain the popularity of a service featuring only short messages. We introduce a new explanatory model for information propagation through social networks that includes selectivity of message consumption depending on their content, competition for user's attention between messages and message content adaptivity through user-introduced changes. Our agent-based simulations indicate that the model displays inherent power-law distribution of number of shares for different messages and that the popular messages are very short. The adaptivity of messages increases the popularity of already popular messages, provided the users are neither too selective nor too accommodating. The distribution of message variants popularity also follows a power-law found in real information cascades. The observed behavior is robust against model parameter changes and differences of network topology.