Research On Domain Ontology Concepts And Relations Learning Algorithm

According to Gruber, Ontologies are formal and explicit specifications of a conceptmodel.As the common understandings of the knowledge of some domain,Ontologies canrepresent the knowledge network of human. Because of their semanticcharacteristic,ontologies can be used in many different occasions,such as informationretrieval,semantic web,knowledge engineering and so on. Thus how to construct apractical ontology has become a hot research topic.Constructing a ontology through human hand is laborious and time-consuming,so it’sgo against the characteristic of large-scale and can be updated dynamically ofOntologies.To improve the automaticity of constructing ontologies has become soimportant that it can never be over emphasized.Ontology Learning,take advantage ofdifferent discipline such as machine learning,data mining,natural lauguage process and soon, aim to address this problem.In this thesis,we propose a series of learning method ofconcepts and relations of one ontology,and organize all these methods into a ontologylearning framework.First,we use two-level TFIDF method to learn domain simpleconcepts, then rule-matching and string frequency is used to learn domain compoundconcepts,we use context dependency method to screen all these learned conceptsafterward.To learn the relations between domain concepts,we first organize all the learnedconcepts hierarchically through extended suffix tree method,then use the combinedmethod of wikipedia link network and hierarchical clustering to improve thisstructure.Methods based on wikipedia category and infobox information can be used toextract domain relations of any kind. Then we use pattern-matching methods to learn morerelation instances with these relations as seeds.The prototype system based on the method mentioned above,not only take fulladvantage of wikipedia,but also assure the scale and timeliness,and is proved itseffectiveness through the experiment conducted.