Acquisition of an aviation safety taxonomy from incident reports and its evaluation

Ontology induction from unstructured text is a long standing goal of Natural Language Processing (NLP). An ontology can serve a number of tasks, such as question answering, Information Retrieval to name a few. In fact, practice communities are often relying on an ontology in their daily decision making process. This is the case of IATA, the partner in this proposal, which currently manually labels aviation incident reports into nodes of an in-house ontology. This time-consuming task is typically conducted by experts. One problem with relying with an ontology is the ontology itself which should evolve over time (e.g., new types of incidents) and is often fraud with inconsistencies. The ontology at IATA is no exception and the goal of the project is to evaluate existing extraction technologies in order to see how useful would an automatically extracted ontology be. One issue with such an endeavor is the evaluation of an ontology. This is typically conducted by asking experts to conduct a manual evaluation. This is at the very least a tedious exercise that definitely prevents the optimization of the acquisition process. Therefore, in this proposal, we also address the issue of automatically evaluating an ontology.

Faculty Supervisor:

Philippe Langlais

Student:

Partner:

International Air Transport Association

Discipline:

Computer science

Sector:

Other services (except public administration); Professional, scientific and technical services; Transportation and warehousing

University:

Université de Montréal

Program: