Application of Natural Language Processing Techniques to Marine V-22 Maintenance Data for Populating a CBM-Oriented Database

摘要

This paper discusses modifications made to the Natural Language Toolkit, a well‐knownrnnatural language processing software package, to achieve improved information extractionrnresults when applied to helicopter maintenance records. In doing so, it will also attempt tornelaborate the components of a tool under development to allow for machine analysis of the freetextrnfields of V‐22 Osprey maintenance records. The authors have found that by adaptingrnexisting natural language processing software to suit peculiarities of the language found inrnmaintenance records, substantive improvements can be made in the accuracy of informationrnextraction. In particular, by modifying an existing text pre‐processor to 1) take in multiplesentencerninputs, 2) treat all code tokens as the same, and 3) ignore distinctions in punctuation,rnpart‐of‐speech tagging accuracy has improved from 92.49% to 96.59%; subsequently, entityrnchunking precision has improved from 91.5% to 92.3%.