New perspectives on intelligence collection and processing

Abstract

Intelligence-production activities are typically viewed as part of an intelligence cycle, consisting of planning, collection, processing, analysis, and dissemination stages. Once a request for information is issued, the intelligence agencies mostly deal with the collection and processing activities of the cycle. However, in most situations, there is an enormous amount of data to be collected. This overabundance of information requires methods that select only the useful data, to prevent intelligence personnel from wasting time and effort on non-relevant data. Online learning is an area of research that has gained attention in recent years with applications in areas such as web advertising, classification, and decision making. In this thesis, we develop a model aimed at the collection and processing phases of the intelligence cycle, applicable in situations where the data is obtained sequentially, so that learning algorithms are realistic. We analyze the performance of a modified Thompson Sampling algorithm, to help intelligence analysts make good decisions, regarding the sources from which to collect/process as well as the collection/processing capacity and its allocation over time, in order to bind the risk of missing valuable information below a certain threshold.