Agent-Based Vector-Label Propagation for Explaining Social Network Structures

Even though Social Network Analysis is quite helpful in studying the structural properties of interconnected systems, real-world networks reveal much more hidden characteristics from interacting domain-specific features. In this study, we designed an Agent-based Vector-label PRopagation Algorithm (AVPRA), which captures both structural properties and domain-specific features of a given network by assigning vectors of features to constituting agents. Experimental analysis proves that our algorithm is accurate in revealing the structural properties of a network in an explainable fashion. Furthermore, the resulting vector-labels are suitable for downstream machine learning tasks.