Networks of interdisciplinary teams, biological interactions as well as food webs are examples of networks that are shaped by complementarity principles: connections in these networks are preferentially established between nodes with complementary properties. We propose a geometric framework for complementarity-driven networks. In doing so we first argue that traditional geometric representations, e.g., embeddings of networks into latent metric spaces, are not applicable to complementarity-driven networks due to the contradiction between the triangle inequality in latent metric spaces and the non-transitivity of complementarity. We then propose the cross-geometric representation for these complementarity-driven networks and demonstrate that this representation (i) follows naturally from the complementarity rule, (ii) is consistent with the metric property of the latent space, (iii) reproduces structural properties of real complementarity-driven networks, if the latent space is the hyperbolic disk, and (iv) allows for prediction of missing links in complementarity-driven networks with accuracy surpassing existing similarity-based methods. The proposed framework challenges social network analysis intuition and tools that are routinely applied to complementarity-driven networks and offers new avenues towards descriptive and prescriptive analysis of systems in science of science and biomedicine.
Download