Automatic Extraction Of Definition Of Terms On Aircraft And Its Security

With the development of the domestic aviation industry ,the training needs for the employees of civil aviation grow rapidly. At present,various training programs and information for assessment is mainly derived from a variety of technical documentation and maintenance manuals,but the training materials for specific purposes and the assessment questions must be repared by technicians or trainers.To solve this issue and achieve the objective of training courseware and assessment questions assisted by computer, this paper processes the corpus on the field of aircraft and its security ,and get the definitions of terms,to provide knowledge for training courseware and assessment questions.To extract the definitions of terms from the corpus on the field of aircraft and its security ,this paper carries out the following work: First, selects, collects and labels the textbooks on the corpus on the field of aircraft and its security ,establishes a corpus, provides the basis for future research. Second, bases on the analysis of linguistic feature of term definition on the corpus on the field of aircraft and its security , and improvements of the patterns present by Zhang Rong and others, Summarizes eight patterns to match and five patterns to exclude,does experiment after compiling the regular expressions according to the patterns, obtains 79.98% of the recall. Then,combines the characteristic of textbooks,first presents the method of term first appearing and experiment, obtains 39.94% of the recall and 16.49% of the precision, although the method when used alone can not meets practical requirements, but when we combines this method with pattern matching,the recall can increased to 88.33%. Then, according to the text clustering ideas, adopts CHI and IG as the feature selection method, works in the platform of Weka use EM algorithm over small data set and big data set for clustering experiments.The results show that when we work at small data set,we can get the best result—68.71% of the MacroF1 and 72.28% of the MacroF2.When we work at big data set,we can get the best result—65.15% of the MacroF1 and 64.90% of the MacroF2.At last,selects all of the standard definition in the corpus,construct a ontology library,provides the knowledge of terms,the relationships between terms ,the definitions of terms and the types of the terms definition for the construction of the ontology library on the aircraft and its security.