Automatic Construction of Ontology

点击翻译按钮获取中文摘要

Ontology as a vocabulary system is important for natural language processing. Constructing the ontology generally requires a great effort. Especially, the ontology about science and technology is difficult to make, since the vocabulary increases continuously according to advances in science and technology. Our goal is to give a procedure for automatically constructing the ontology about science and technology.At first we gave a mathematical basis of the ontology. We defined the ontology to be an ordered set, and introduced ontology into it. We showed some mathematical properties on the ontology, and developed a procedure to build the ontology from the observed collection of open sets.Next, we tried to derive the ontology from the sentences of the abstracts in INSPEC tapes. We considered as open sets the sets of subjects which takes the same attribute-adjective, and extracted from the sentences of the form : 'Subject+be+attribute-adjective'. However, thereare few pairs of sets which are … More in the inclusion-relation. Then, we extracted pairs of subjects and nouns as the ISA-relation directly from the sentences of the form : 'Subject+be+article+noum'. The resulting ontology contains a relatively long sequences of nouns.These experiments tells that the problem is the lacking of the common sense knowledge in the abstracts. One can complement such knowledge by using existing ontology, such as WordNet. However, it is doubtful that the general purpose ontology is useful as the common sense knowledge in the scientific and technical documents. In the sentences that form the pattern 'compare A with B', the noun phrases A and B are 'comparable'. Therefore we assumed that the head nouns in A and B are 'similar' in the sense that they have a common ancestor in the ontology. Experimental results using WordNet shows that 75% of the pairs extracted from the sentences have a common ancestor in WordNet. The reason why the 25% pairs are not on the ontology is mainly due to the fact that the words such as 'data', 'result'are frequently omitted. Less