Assessing the value of localized centrality metrics

Acquiring the full global information is impractical, if feasible at all, in many networks with distributed operation and self-organization features. To meet scalability requirements practical protocol implementations could use local information instead, drawn from the nodes’ ego-networks, the Social Network Analysis (SNA) counterpart of centered graphs. However, in almost all these efforts the capacity of local ego-centered measurements to substitute global information is not evaluated but rather taken for granted. It is this assumption that we effectively put under the microscope. We focus on network centrality metrics and explore to what extent sociocentric metrics determined under complete network-wide information (rank-)correlate with their ego network counterparts. We generalize the definition of ego network, to comprise the ego’s r-hop neighborhood, respectively, and analyze the involved accuracy vs. complexity tradeoff. The correlation of the global and ego metric variants is found high in synthetic and even higher in real-world network topologies of several hundred nodes size. We discuss the impact of network topology on this correlation and possible network functions that can benefit from rank-preserving ego-centered measurements. Our findings can serve as evidence for the practical feasibility of “socioaware” networking.