Data sources relevant to homeland security are diverse, both in content and format, so it becomes imperative to define a uniform data model that enables information extraction. Our project explores the use of a "Universal Information Graph" (UIG) obtained by "virtual fusion" of Specialized Infrastructure Graphs (SIG's) such as the Web, the Internet, Call Detail, Email and the Airline Transportation Network. The information associated with each edge in a SIG comes from URL accesses, data, voice and text messages, travelers' lists, etc. Information flow patterns constitute one of the intrinsic characteristics of how a particular infrastructure is used by like-minded individuals to perform a task or achieve a goal. The challenge is to understand how these basic information flow patterns persist when their underlying graphs get "virtually fused" into the UIG (rather than actually fusing them since this would create a UIG that is too massive and too unwieldy). Our approach builds on the notion of node importance used in Web search ranking and social networks research. We investigate computationally efficient node rank functions and similarity measures for homeland security applications that can be implemented in distributed computing architectures. We also study the foundational question of how to approximate various linkage metrics with limited time and space resources.
Document last modified on August 17, 2007.