Research On Text Categorization Methods And Its Application To Analysis Of Civil Aviation Safety Reports

Aviation safety is a big problem in the industry of civil aviation. In order to enhance civil aviation system, detect and correct system deficiencies as quickly and accurately as possible, control and eliminate hidden dangers to aviation safety, countries around the world have set up aviation safety reporting systems to collect reports of unsafe incidents and potential safety problems.Existing aviation safety report systems contains the analysis of structural data from those reports most of which are simple data queries and statistics. The analysis of unstructured data,which requires hard works form domain experts, is still a challenge. Artificial analysis is time-consuming and labor-intensive; the accuracy of analysis depends a lot on staff capacity and experience. And as time goes on, reports accumulate rapidly; the quantity is now far beyond the capacity of experts. Therefore, the study of effective analysis methods, discovering the unknown and potential problem automatically and accurately, revealing the events related to the airline control and eliminating potential safety problems has an extremely important practical significance to improve aviation safety performance.In this paper, the technique of automatic text categorization is applied to the analysis of a typical kind of aviation safety report, the ASRS report. Works in this paper are mostly focused on the following topics:1? A brief summary of the development of text categorization technique and aviation safety reporting systems is given, the need of employing text mining techniques to analyze aviation safety reports is explained.2? A fast and effective text categorization method based on local document frequency is proposed after the study of existing methods.3? Several text categorization methods are applied to categorize ASRS reports, the analysis of several problems encountered during application.4? A fast and effective text categorization system is developed in order to fulfill the need of a platform specified for text categorization of large amount of data.