Text-oriented Ontology Learning Research In The Concept Extraction And Relation Extraction

Research on ontology is becoming increasingly widespread in the computer science community. But the manual construction of ontology is a time-consuming task and easily leads to the bottleneck of knowledge acquisition. Ontology learning aims at constructing ontology semi -automatically by integrating a multitude of disciplines such as ontology engineering and machine learning. Ontology learning refers to the extraction of learning content (concept knowledge) and uses this content to construct ontology. There are many different types of data in the real world such as text, XML, HTML and DTD. Most of the data can be used as the data sources of ontology learning. There are different ontology learning approaches according to the type of input: ontology learning from structured-data, from semi-structured data and from unstructured data.In this paper, both theory and experiment researches have been performed on the concept extraction and relation extraction. So several aspects are mainly researched and new ideas and innovations have been made as follows:Make a systematic and profound study of the basic theories of ontology including its definition, character, type, described language and methods of constructing ontology.A method of ontology learning cycle has been researched. This paper especially introduces the learning content, the existent ontology learning system, the system architecture which is now widely used in ontology learning and evaluation of ontology. Features of the four most representative ontology learning systems are discussed first in this paper.The new ontology learning system from text (OLSFT) has been proposed and this system includes ontology management component, resource processing component, algorithm library component and coordination component. This paper mainly introduces the concept learning algorithm and relation learning algorithm. The statistic approach is used in the algorithm of concept extraction and we improve on the general statistic approach based on the assumption. By using domain relevance and domain consistent the concept extraction can be more representative. The extraction of concept relations mainly includes the following approaches: lexico-syntactic patterns and association rules. What's more, the future direction of improving the algorithm is described.Also, all the necessary modules of OLSFT system are discussed in this paper including the concept extraction module and the relation extraction module. According to the different modules different performance indicators are brought forward. All of these are useful for constructing ontology learning system. The experiments have proved that the innovations in this paper are feasible.