AbstractThe Railway Industry in Great Britain is on the brink of a seed change in the methods used to carry out safety and risk management. Advances in technology over the past few decades have enabled the collection of varied and detailed information regarding the state of track, vehicles, stations, facilities, and personnel at an ever-increasing scale. As more and more industry processes embrace data driven techniques, the growing volume of data being recorded poses a significant challenge in terms of processing and knowledge extraction.
In these chapters, a case study is made of the Close Call system used by railway organisations in Great Britain. Personnel working on the railway network can make reports to this system by phone, email, and in-app, if they see something they consider to have the potential to cause harm or damage. The near ubiquity of mobile devices allows all staff to make free-text reports at any time which has great potential for the completeness of reporting but comes with some drawbacks. The freetext nature of the reports means that basic processing is inadequate for tasks such as categorisation, knowledge extraction, and reporter feedback, requiring ongoing research into appropriate natural language processing techniques. A variety of techniques have shown promise, but the vast and growing quantity of close call reports being made has meant that computing capacity has limited the number of reports that can be processed, and work so far has often been applied only to a small subset of reports.
The aim of this thesis is to demonstrate that it is possible to process or otherwise transform the close call text to allow existing and future analysis techniques pertaining to this text (and data sources like it) to be applied on a larger scale or smaller timeframe. A novel text indexing technique is presented and evaluated, and when compared to other text indexing and pattern matching methods, it is shown that a significant speed increase is possible over the methods previously used.
|Date of Award||2023|
|Supervisor||Coen Van Gulijk (Main Supervisor) & Violeta Holmes (Co-Supervisor)|