: 

Intelligence Collection in a Glut of Information by Yuval Nevo

Yuval completed his masters thesis at NPS in the winter of 2011.   His work concentrated on intelligence collection in the presence of an overwhelming amount of information.  Below you will find his executive summary, and the thesis itself.

Executive Summary

One of the key stages in producing intelligence is Processing and Exploitation. Within this stage, the collected raw data is transformed into usable information. In modern intelligence agencies, one of the main obstacles in the Processing and Exploitation stage is the abundance of information, which makes differentiating between relevant and irrelevant information a difficult task. Due to time constraints, an intelligence processor of collected raw data, called henceforth a collector, cannot process all the collected intelligence items and therefore some screening procedure is needed. In this research we address the information selection problem: Which intelligence items should the collector screen in order to maximize the expected amount of relevant information screened?

The information selection problem can be seen as a part of a broader class of problems called exploration-exploitation problems. In an exploration-exploitation problem one has to repeatedly choose between several alternatives, and faces the tradeoff between exploring (investigating new alternatives) and exploiting (utilizing familiar alternatives). The information selection problem has unique characteristics, making it a relatively difficult exploration-exploitation problem. Specifically, intelligence sources are dependent; the information gained from the screening process of one source can be used to better estimate the relevance value of other sources.

In order to handle the information selection problem, we develop a mathematical model of the information screening process. The model handles a situation in which a collector faces a pool of intercepted conversations, which he needs to screen. We explored several selection algorithms that would allow the collector to detect as many relevant information items as possible. Based on the mathematical model, we constructed a simulation of the screening process. We then examined the performances of several selection algorithms, using a scenario based on the terrorist network behind the U.S. embassy attack in Tanzania in 2007.

The main contributions of the thesis are the mathematical model of the screening process, the selection algorithms and several important insights detailed below:

  • Simple selection algorithms, which we examined, performed much better than anticipated. We anticipated that a simple greedy algorithm and another basic algorithm called “Softmax” would perform much worse than more advanced algorithms. However, the performance of these algorithms was quite well compared to the advanced algorithms. We speculate that the dependencies among the alternatives are the main reason for that performance.
  • The algorithms which showed the best performance are an algorithm based on the Knowledge-Gradient policy and an intuitive heuristic for screening the conversations. The Knowledge-Gradient policy is an exploration method in which one chooses the alternative that is most likely to change its beliefs regarding the value of the different alternatives.
  • The mean number of conversations between the different persons is a significant factor in the performance of the algorithms. When the mean number of conversations is small, there is no significant difference between the performances of the different algorithms.

Thesis Presentation

Complete Thesis

Nevo, Y. K. (2011).  Information Selection in Intelligence Processing (Masters Thesis).  Naval Postgraduate School.

© Copyright 2004-2012 - Ned Dimitrov