Network Extraction and Resolution System (NEaRS)

The increasingly asymmetric nature of U.S. military operations has resulted in a need for Warfighters to understand the"human terrain"in which they operate. To support tactical decision-making, a social network should be constructed from the widest array of available intelligence sources, such as TIGR, BAT/HIIDES, and OSINT. Producing the social network from this variety of sources presents two major challenges: First, these sources use a number of different formats and modalities, are often unstructured or poorly formatted, and do not share a common representation. Second, the sources each individually provide noisy and partial information, which must be combined with information from other sources to produce a common picture. To address these challenges, Charles River Analytics proposes a Network Extraction and Resolution System (NEaRS) that will use multi-source graph resolution (MSGR) to estimate the true social network from noisy observations from multiple sources. MSGR will build on insights from statistical relational learning to model uncertainty in the sources, use relational information, and fuse information from diverse sources, while scaling to large networks. We will also extend MSGR to dynamic graph resolution to allow the social network estimate to adapt to dynamic developments.