Paper accepted for the 2005 Intelligence Analysis Conference, May 2005, Washington, DC. Information Triage with TRIST David Jonker, William Wright, David Schroh, Pascale Proulx, Brian Cort Oculus Info Inc. {david.jonker, bill.wright, david.schroh, pascale.proulx, brian.cort}@oculusinfo.com Keywords: Multi-INT/fusion, All Source Intelligence, Novel Intelligence from Massive Data, Search and Retrieval Abstract TRIST (“The Rapid Information Scanning Tool”) is the information retrieval and triage component for the analytical environment called nSpace. TRIST uses Human Information Interaction (HII) techniques to interact with massive data in order to quickly uncover the relevant, novel and unexpected. TRIST provides query planning, rapid scanning over thousands of search results in one display, and includes multiple linked dimensions for result characterization and correlation. It also forms a cohesive platform for integrating computational linguistic capabilities such as entity extraction, document clustering and other new techniques. Analysts work with TRIST to triage their massive data and to extract information into the Sandbox evidence marshalling environment. Initial experiments with TRIST show that analyst work product quality is increased, in half the time, while reading double the documents. 1 Introduction As part of the Novel Intelligence in Massive Data (NIMD) research program [ARDA, 2002], new interactive, information visualization techniques are being investigated which tightly couple massive data, software agents and the analyst’s exploration task. A break- through in finding novel intelligence is believed possible if all the components are combined in a system of systems. Progress has been made towards an integrated cognitive space where analysts will see, and interact with, more information, more quickly, with more comprehension. This space is called "nSpace" and is the combination of the multi-dimensional linked views found in TRIST with the visible and flexible cognitive mechanisms of the Sandbox. Figure 1. Workflows supported by nSpace. This paper focuses on TRIST. First, the results of a cognitive task analysis are presented. This is followed by a discussion of related information visualization work. Then there is a technical discussion of TRIST concepts and capabilities. Finally, results are reviewed from an experiment conducted at NIST that measured the performance impacts of TRIST. 2 Analysis Cognitive Task Analysis Cognitive task analysis (CTA) is a layered system-based framework used to optimize the match between a task’s cognitive demands and the design solution [Schraagen, 2000]. CTA techniques include a variety of systematic observation, interview and analysis methods, all aimed at providing insight into the mental processes underlying complex human tasks. In preparation for investigating nSpace concepts, two CTA studies were completed: structured interviews and a review of analyst activity as logged by the NIMD “Glass Box” [Cowley et al, 2004]. The observations from these CTA studies were used to refine the performance objectives for TRIST and potential task metrics. 2.1 Speaking with Analysts - Observations Structured interviews were conducted with fourteen analysts who work with a variety of sources and on a range of short/long term, narrow/broad focus subjects [Wright and Kapler, 2004]. The interviews were wide ranging and touched on many topics: IR, Analysis Methods, Tools, Work Products, Objectives, etc. With respect to information retrieval, it is clear that executing queries, sequentially scanning results, opening and reading documents is a common, time consuming task for analysts. The following are illustrative excerpts on massive data and IR: “You start with 30,000 hits … which you need to understand and put in some order. And I need to see documents and in nine different ways. “I have about 1,000 messages or reports a day. About 300 are relevant. And maybe 30 go to my log. I cut and paste fragments into my log. Doing updates takes most of my time. “I need to see the searches. I need to jump back to any level in the searches. … And I can’t refine searches now …and there are no nested searches … © 2005 Oculus Info Inc. TRIST - The Rapid Information Scanning Tool Page 1