7/
Kudos to Juan Castillo
check the preprint here 👉 arxiv.org/abs/2510.14098

6/ Policy implication: early, proactive interventions matter. To unlock a silenced group, it’s not enough to “undo” a setting; we may need targeted rewiring and the right behavioural noise to cross basins of attraction.

5/ Why platforms should care: algorithms that intensify triadic closure can tip discourse into a silent-minority regime that doesn’t recover just by dialing the knob back. The return path is different—classic hysteresis.

4/ Nuance: moderate clustering protects minority expression by creating supportive local pockets; excessive clustering entrenches asymmetries (homophily amplification → core–periphery), making silence stickier

3/ Monte Carlo confirms it: the hysteresis loop persists in the stochastic system, narrower than mean-field, but still there. Structural (irregular topology, core–periphery patches) and behavioural noise helps the system escape or traps in some metastable states

2/ Mechanism in a nutshell: as triadic closure + choice homophily shape exposure, the expression dynamics undergo saddle bifurcations, creating multiple attractors. That’s the seed of path dependence and history-locked outcomes.

These oscillations offer a practical diagnostic of endogenous criticality in social contagion.
We test the method on real Twitter data distinguishing intrinsic transmissibility shifts from external shocks.

Kudos to Eva Rifá and Julian Vicens

🔗 arxiv.org/abs/2505.24795

We show that:
🔹 Targeted spreaders can extend or shorten rumor lifetime
🔹 Autocorrelation acts as an early warning signal near criticality
🔹 Only endogenous growth yields oscillatory autocorrelation (captured analytically via NIMFA)