IMPROVING OPERATIONAL REPORTING WITH ARTIFICIAL INTELLIGENCE

Abstract

Today, military analysts receive far more information than they can process in the time available for mission planning or decision-making. Operational demands have outpaced the analytical capacity of the Department of Defense. To address this problem, this work applies natural language processing to cluster reports based on the topics they contain, provides automatic text summarizations, and then demonstrates a prototype of a system that uses graph theory to visualize the results. The major findings reveal that the cosine similarity algorithm applied to vector-based models of documents produced statistically significant predictions of document similarity; the Term Frequency-Inverse Document Frequency algorithm improved similarity algorithm performance and produced topic models as document summaries; and a high degree of analytic efficiency was achieved using visualizations based on centrality measures and graph theory. From these results, one can see that clustering reports based on semantic similarity offers substantial advantages over current analytical procedures, which rely on manual reading of individual reports. On this basis, this thesis provides a prototype of a system to improve the utility of operational reporting as well as an analytical framework that can assist in the development of future capabilities for military planning and decision-making.