Learning Social Networks from Text Data using Covariate Information

Abstract

Describing and characterizing the impact of historical figures can be challenging, but unraveling their social structures perhaps even more so. Historical social network analysis methods can help and may also illuminate people who have been overlooked by historians but turn out to be influential social connection points. Text data, such as biographies, can be a useful source of information about the structure of historical social networks but can also introduce challenges in identifying links. The Local Poisson Graphical Lasso model leverages the number of co-mentions in the text to measure relationships between people and uses a conditional independence structure to model a social network. This structure will reduce the tendency to overstate the relationship between "friends of friends", but given the historical high frequency of common names, without additional distinguishing information, we can still introduce incorrect links. In this work, we extend the Local Poisson Graphical Lasso model with a (multiple) penalty structure that incorporates covariates giving increased link probabilities to people with shared covariate information. We propose both greedy and Bayesian approaches to estimate the penalty parameters. We present results on data simulated with characteristics of historical networks and show that this type of penalty structure can improve network recovery as measured by precision and recall. We also illustrate the approach on biographical data of individuals who lived in early modern Britain, targeting the period from 1500 to 1575.