Adaptive edge weights reshape opinion dynamics on networks

A team led by Mason A. Porter at the University of California, Los Angeles (UCLA) has published a paper extending the well-known Deffuant-Weisbuch (DW) bounded-confidence model of opinion dynamics by introducing adaptive edge weights that govern interaction probabilities between agents. In the classic DW model, agents only influence each other when their opinions are sufficiently close (within a confidence bound). The new work, posted on arXiv (2605.20418), adds a layer of realism: the probability of two agents interacting evolves over time based on their history of successful compromises or other positive exchanges. The researchers prove theoretical guarantees for the model's convergence, show how edge weights change over long timescales, and define an 'effective graph' — a time-dependent subgraph including only currently receptive agent pairs.

Numerical simulations on a variety of network topologies reveal a nuanced picture. For small confidence bounds, adaptive edge weights accelerate consensus formation in dense networks (more connections) but actually slow it down in sparse networks (fewer connections). This suggests that the structure of social ties and the memory of past interactions together shape how quickly groups reach agreement or polarize. The work bridges statistical physics, network science, and social dynamics, with potential applications in modeling online echo chambers, political polarization, and the spread of ideas in communities where people preferentially communicate with those they've previously aligned with.