Collaboration mechanisms and community detection of statisticians based on ERGMs and kNN-walktrap

Abstract

Comprehensive information on coauthorship from 2014 to 2018 was gathered from four top statistical journals and subsequently cleaned to provide a review in the field from the perspective of a co-authorship network analysis. Data on productivity and trends, as well as a skew analysis of publications and collaborations, was provided by the analysis. The coauthorship network was analyzed for both global and individual properties. Exponential random graph models (ERGMs) were also used to explore the formation mechanisms of collaboration while simultaneously considering exogenous covariate effects and endogenous network structure processes. It was discovered that homophily (authors from the same universities and countries) and transitivity (the tendency to collaborate with a coauthor's coauthor) have a significant positive effect on the production of collaborative studies. Finally, the kNN-walktrap was proposed, which combines the structures of the network and the homophily features of authors to detect network communities. In this method, the cosine similarity calculated by the homophily features of the nodes is utilized to build a kNN (k Nearest Neighbor) network and apply walktrap to detect communities. Thus, more detailed and comprehensive community structures can be detected than when using the walktrap method. These results have practical significance for researching collaboration models and guiding future collaboration.