Statistical Modelling of Citation Patterns for Publications in Statistics Journals

Abstract

The main aim of our computational project is to understand the topology of the community of statistical research. Citation data of scientific articles published by 7 statistical journals are cleaned. We construct a citation network with the papers as nodes and a total of 7 communities are found by spectral clusterings. We further construct covariates from our dataset to the nodes in the network. A novel variational inference method on community recovery of stochastic blockmodels is developed by incorporating nodal informations. The new likelihood method is implemented on our network, we compare our results with the ones from the nonparametric spectral clustering. Their empirical differences in community recovery are examined, and we show evidence that the results from our variational approach are equally meaningful and are of consequences from using the covariates.