Distances in weighted higher-order networks

Abstract The concept of distance is a fundamental idea in graphs and hypergraphs. However, its extension to weighted hypergraphs is challenging, since it may result in inconsistencies, especially if the weights are arbitrarily assigned to the hyperedges. We address this challenge by proposing a general distance measure for weighted hypergraphs. Our measure is well-defined, and it reduces to the classic graph distance if the edges in the hypergraph are all pairwise links. We demonstrate the applicability of our definition by analyzing a number of real-world higher-order datasets, including the network of preprints in the arXiv repository, for which we choose the weights in a way that reflects the notion of cognitive distance, which measures the conceptual remoteness between fields. Our results demonstrate that, when higher-order edges cannot be neglected, the use of a full hypergraph measure is necessary to avoid the information loss that would result from commonly used approaches, such as clique projection.