To investigate the relationship between local atomic structures and particle dynamics in metallic glass-forming liquids, we examine the connection between particle relaxation behavior and the global topology of transient network structures. Local atomic structures are identified using Voronoi polyhedra, and the system is mapped onto an unweighted, undirected graph, where atoms are represented as vertices and atoms with the same local structure are connected when they are nearest neighbors. The local connectivity is used to characterize the number of connections of an atom, and the self-intermediate scattering function is calculated to distinguish the relaxation behavior of particles with different connectivity. The results show that, in Cu50Zr50, the relaxation behavior of 〈0,0,12,0〉 icosahedral clusters depends distinctly on connectivity, whereas 〈0,2,8,2〉 and 〈0,3,6,4〉 clusters exhibit weak dependence on connectivity. In Ni50Al50, the self-intermediate scattering functions of 〈0,0,12,0〉 clusters nearly overlap for different connectivity levels, indicating weaker dynamic specificity. Network analysis further shows that the transient networks deviate from the Erdős–Rényi random-network model even when the local connectivity is close to the random distribution. These results demonstrate a statistical connection between particle relaxation behavior and the global topological features of transient network structures, establishing a graph-dynamics correspondence in metallic glass-forming liquids.