Estimating node similarity from co-citation in a spatial graph model

Co-citation (number of nodes linking to both of a given pair of nodes) is often used heuristically to judge similarity between nodes in a complex network. We investigate the relation between node similarity and co-citation in the context of the Spatial Preferred Attachment (SPA) model. The SPA model is a spatial model, where nodes live in a metric space, and nodes that are close together in space are considered similar, and are more likely to link to one another. Theoretical analysis of the SPA model leads to a measure to estimate spatial distance from the link information, based on co-citation as well as the degrees of both nodes. Simulation results show this measure to be highly accurate in predicting the actual spatial distance.