A Quantitative Methodology for Vetting Dark Network Intelligence Sources for Social Network Analysis

Social network analysis (SNA) is used by the DoD to describe and analyze social networks, leading to recommendations for operational decisions. However, social network models are constructed from various information sources of indeterminate reliability. Inclusion of unreliable information can lead to incorrect models resulting in flawed analysis and decisions. This research develops a methodology to assist the analyst by quantitatively identifying and categorizing information sources so that determinations on including or excluding provided data can be made. This research pursued three main thrusts. It consolidated binary similarity measures to determine social network information sources\u27 concordance and developed a methodology to select suitable measures dependent upon application considerations. A methodology was developed to assess the validity of individual sources of social network data. This methodology utilized source pairwise comparisons to measure information sources\u27 concordance and a weighting schema to account for sources\u27 unique perspectives of the underlying social network. Finally, the developed methodology was tested over a variety of generated networks with varying parameters in a design of experiments paradigm (DOE). Various factors relevant to conditions faced by SNA analysts potentially employing this methodology were examined. The DOE was comprised of a 24 full factorial design augmented with a nearly orthogonal Latin hypercube. A linear model was constructed using quantile regression to mitigate the non-normality of the error terms