Fig. 1.

Performance on synthetic networks as a function of time horizon and class dynamics, as measured by the adjusted Rand index between true and estimated community labels. Networks were generated under a dynamic DCBM (Eqs. 13–15) with three key parameters: $p_{\text{in}}$ and $p_{\text{out}}$, which determine the in-cluster and out-of-cluster edge probability/density, and $r$, which determines the amount of change in cluster memberships between consecutive networks (0 for no change). For A, $K=2,n=100,p_{\text{in}}=(0.2,0.25),p_{\text{out}}=0.1$, and for B, $K=10,n=500,p_{\text{in}}=(0.2,0.35),p_{\text{out}}=0.1$. Shown are 100 simulations per data point.