ROUGH DRAFT Concept Map Value Propagation for Tactical Intelligence

Command and control analysts increasingly apply traditionally unassociated concepts to situation understanding. Techniques are needed to tailor knowledge acquisition resource allocation according to probable value of information, both inferring an answer to a question using knowledge at hand and quickening intelligence efforts to fill in gaps. This paper discusses research into techniques for negotiating and updating Concept Maps (Cmaps), collections of concepts linked by specified relationships, by accounting for both nodal data and links among nodes. Such Cmaps provide the flexibility to represent contemporary tactical knowledge not lending itself to conventional data structures. In some senses Concept Mapping generalizes the notion of an inference network, a set of propositions organized with rules directing information propagation and combining antecedents to update consequents. We are attempting to develop a mathematical system for organized navigation of a Concept Map, driven by expected variability in the value of a datum and cost to get a new value. We use the CmapTools software associated with the Institute of Human and Machine Cognition as a structural basis for creating and assessing tactical Cmaps. The paper discusses Cmap construction, analytical philosophy, and methodology development. 1. Background and motivation With the rise of net-centric and effects-based operations in response to asymmetric warfare, analysts have begun to apply traditionally unassociated concepts to situation understanding and intelligence gathering. Dynamic network analysis tools are used to locate and assess networks embedded within large data sets and to establish exploitable patterns. Several such tools exist; however, they are generally inadequate in that analysts spend too much time accomplishing “mechanical” tasks that severely reduces time remaining for analysis. Techniques that focus data extraction and knowledge acquisition in accordance with the probable value of the information will alleviate the analysts’ burden. We envisage an effort that complements context-based enhancement of network analysis. We will capitalize on associated work in text mining that produces node/relation attributes (e.g., for socio-cultural and spatio-temporal information). Concept Maps (or Cmaps), having generalized graph structures and “arbitrary” labels, present problems beyond those of conventional structures used for inferencing. Their strength, unlimited semantics, appears initially to be a weakness from the standpoint of computability. However, we are attempting in this project to address this issue. To improve knowledge acquisition an intelligence tool should combine ease of answering with influence on the user-specified top concept. -We wish to develop a system to compute the next task in an intelligence mission, where the determination is based on expected variability in the value of existing data per unit cost to get a new value. Such functions represent the ability of an antecedent to change the value of a direct consequent. A high value of such a function indicates that a node influences the top concept relatively greatly with relatively low cost of obtaining additional information about the node. A decision maker is typically unable to quickly provide reliable estimates for all aspects of a situation. However, he or she can specify (say, via Cmaps) beforehand items the decision depends on. These depend on others in turn, and this configuration is specified as a network that propagates information. We wish to develop in mathematical terms qualities that enable the next proposition to be expanded when dealing with values in a concept network. We are investigating ways to express values in a Concept Map in terms of the value of a top consequent -divided by cost. Leading to a control strategy utilizable with mixed types of propositions, this work involves considerations of the value of information and derivation of influence. This effort involves three linked aspects: construction of tactical Cmaps; development of the analysis methodology; and feasibility assessment. We are exploring the use of Concept Maps to describe tactical situations and facilitate military tactical decision making. Further, we intend to implement techniques for negotiating and updating military situational Concept Maps by attempting to take into account influences and values of relationships among nodes. This latter objective is potentially of great value to the intelligence analyst in tailoring allocation of resources for knowledge acquisition, in the spirit of sensor allocation to obtain commander’s critical information requirements and with knowledge of costs for achieving desired results. Our technical process involves application of team expertise to various subtasks, including: search for suitable cognitive mapping methods and flexible combat simulation; incorporation of social network analysis; output into development of a “scenario” (set of related concepts) for populating the map(s); development of mathematical algorithms for traversing a non-hierarchical array of linked nodes; development of one or more experiment(s) inclusive of metrics, to characterize the value of the prototype; implementation of improvements suggested through experimentation; final prototype evaluation; and testbed integration.