Preliminary Evaluation of an Aviation Safety Thesaurus' Utility for Enhancing Automated Processing of Incident Reports
Francesca A. Barrientos, Joseph P. Castle, Dawn Marie D. McIntosh, Ashok N. Srivastava · 2007
The purpose of this document is to present a preliminary evaluation the utility of the FAA Safety Analytics Thesaurus (SAT) in enhancing automated document processing applications under development at NASA Ames Research Center (ARC). Current development efforts at ARC are described, including overviews of the statistical machine learning techniques that have been investigated. An analysis of opportunities for applying thesaurus knowledge to improving algorithm performance is then presented. Background The Intelligent Data Mining group at NASA Ames Research Center has been developing machine learning algorithms and software tools to perform text mining and other document processing on the Aviation Safety Reporting System (ASRS) and Aviation Safety Action Program (ASAP) incident report databases. Two different problems are being addressed by this effort. The first is the automated categorization (classification) of incident reports by event type. The event types are drawn from the Distributed National ASAP Archive (DNAA) Master List [1] of 31 primary event types, and a report may belong to more than one event type category. The second task is to identify the