Network graph data has multitude of usages. It serves the purpose of identifying and conceptualizing interrelations and dependencies between entities from a large variety of domains including Telecom networks, social network data, GIS, medical data, e-commerce data, collaboration networks, web-links and more. This graph data is later consumed by a variety of end applications for equally varied purposes. In this paper we present an analysis of various methods of computing values of various kinds of centralities such as network simulation, fluid flow simulation, randomized algorithm, and machine learning-based predictions across a sample of graphs varying in sizes from 34 entities to 40,164 entities, with further emphasis on the application of computed centrality information for the determination of deep communities on the graphs under consideration. We also use a method for computing betweenness centrality for graphs based on a randomized algorithm, and analyze it empirically.