Similarity Measures for Human Networks

The search for entities of interest (either individuals or groups) in large behavioral datasets has emerged as a key need for irregular warfare, counter-insurgency, and counter-terrorism. This search requires measures of similarity between networks, and the combination of diverse multi-dimensional measures into a single meaningful scalar measure of similarity. To address this challenge, Aptima has teamed with two leaders in the field of human network research: Dr. Kathleen Carley from Carnegie Mellon University and Dr. Alex (Sandy) Pentland from the MIT Media Lab. Dr. Carley brings a meta-matrix framework developed for computational organizational modeling to the team. Dr. Pentland brings his sociometric network similarity measurement work, which introduces similarity measures based on data from new types of wearable sensors. Aptima’s work on human network similarity from a system engineering and network science perspective forms a third component and serves as a bridge. Testing of measures in Phase I will use a hybrid real-synthetic dataset. A hypothetical test case will represent a “person or group of interest” and the Aptima team will calculate measures of similarity between the entities in the dataset and the synthetic target of interest, analyzing the effects of varying levels of uncertainty on the similarity measure.