Title:Transitive Reduction of Citation Networks

Abstract:Citation networks form directed acyclic graphs. The transitive reduction of such a DAG removes all edges in the graph that are implied by transitivity, but retains the causal structure of the original network. We investigate how the degree distribution, clustering coefficients and relationship between time of publication and degree (representing citation count) are changed under transitive reduction. We show that models which accurately reproduce the degree distribution of a citation network before transitive reduction may not do so afterwards and hence suggest that these models fail to accurately capture the causal structure of the network. We also investigate how the citation count of individual papers is changed and suggest that information about the types of citations a paper receives can be inferred from the number of citations which are retained after transitive reduction. These methods are tested on citation networks from the hep-th and hep-ph sections of the arXiv repository, and all data has been made publicly available.