Research On Ontology Concept Learning And Ontology Relation Learning

With the rapid development of Internet,network resources have shown explosive growth.The over-inflation of information has posed great challenges to knowledge retrieval,management,and application.Ontology,as a kind of knowledge organization method,can effectively represent domain knowledge,so as to achieve various researches at the semantic level,and improve the efficiency of management and acquisition of knowledge.Ontology is mainly composed of two parts:ontology concept and ontology relation,and it is difficult to manually construct ontology,so the automatic extraction of ontology concept and ontology relation has important theoretical and practical application significance.Two ontology learning methods are mainly studied,which are used to learn(semi)automatically the ontology concept and ontology relation from the Chinese domain corpus.The main research contents and conclusions are as follows:(I)Research on ontology concept learning.Ontology concept is an important component of ontology learning,in order to improve the effectiveness of(semi-)automatic construction of domain ontology.This study studies an ontology concept learning method based on domain corpus.The method firstly extracts the candidate term set by the atomic word step method and the information entropy,then uses the domain membership thesaurus to learn the domain specific term set from the candidate term set,then merges synonymous terms,and finally obtains the domain ontology concept set.This method is simple and easy to use,and the information entropy can greatly reduce the proportion of non-words in the candidate term set.It has positive significance for expanding ontology term collections and constructing ontology concept collections(2)Research on ontology relation learning.In order to improve the accuracy of ontology relationship extraction,this study proposes an ontology relationship learning method based on neural network.Firstly,the concept vector model is constructed through Word2Vec model.Then the concept similarity is calculated and used to obtain the concept relation vector model.Next,the convolutional neural network is used to construct the ontology relation classification model,and the classification result is submitted to the domain expert for correction.Finally,the ontology relation set is obtained.This method can accurately and effectively distinguish the relations of concept after learning of manually annotated data sets,therefore it can be extended and used in multiple fields.To verify the validity of the proposed method,the study carries out two experiments on Fudan corpus by the methods of the ontology concepts and ontology relations.The result proves that the ontology concept learning and ontology relationship learning methods proposed in this paper have good performance,and can support automation in the ontology construction process to a certain extent.