AUTOMATIC SECURITY CLASSIFICATION STUDY

An investigation was made of the feasibility of using computers to assign the proper security classification (unclassified, confidential, secret) to textual material. The words in 998 paragraphs were transformed to computer-usable form. A set of 66 variables was computed for each paragraph by a two-...

Full description

Saved in:
Bibliographic Details
Main Authors: Enger, Isadore, Merriman, Guy T, Bussemey, Ann L
Format: Report
Language:English
Subjects:
ALGORITHMS
AUTOMATIC
CLASSIFICATION
COMPUTERS
DATA PROCESSING
FEASIBILITY STUDIES
Information Science
MILITARY PUBLICATIONS
NATIONAL DEFENSE
SECURITY CLASSIFICATIONS
STATISTICAL ANALYSIS
VOCABULARY
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:An investigation was made of the feasibility of using computers to assign the proper security classification (unclassified, confidential, secret) to textual material. The words in 998 paragraphs were transformed to computer-usable form. A set of 66 variables was computed for each paragraph by a two-stage process of attaching three scores to a word and then combining the scores in various ways over the words of a paragraph. Several experiments were conducted to validate assumptions involved in the method of scoring the words and the methods for combining the scores. The 66 variables were presented to a statistical technique which made a preferential selection of a small set of effective variables from the large set of 66 variables. The redundant or non-controlling variables were eliminated from subsequent analysis, and an objective system was developed for assigning security classifications using only the selected variables. The system was applied to an independent sample of paragraphs and 53.9 percent were correctly classified. It was concluded that the system does exhibit skill. However, the skill is probably too low to consider replacing the present system. Finally, it is concluded that the method for forming variables and the statistical technique, both apparently new to this field, show sufficient promise to merit application to other automatic indexing problems.