Research On Key Technologies Of General Ontology Learning Method And Its Application

With the rapid development of network technology and the continuous improvement of social information, the people’s demand of the information system on intelligence and knowledge is increasing day by day in all aspects. So, the representation of data not only stays at the grammatical level, but also needs to focus on the semantic level. Ontology is conceptual model which provides a better method by describing information and data on the semantic level, and provides an effective method to solve the problem about semantic level of understanding and communication. At present, there are two ways in ontology construction. One is manual ontology construction by the tools of ontology construction, and another is ontology learning by constructing ontology automatically or semi-automatically. With the arrival of the era of big data, the way of ontology learning method to construct ontology is becoming more and more important.The existing ontology learning method is basically determined by the structure of the data source of ontology learning. There are different ontology learning methods according to the different structures of data sources. However, it is solved problem that how to extract ontology knowledge from heterogeneous data sources to construct and rich ontology through general ontology learning method in the huge data information on the internet.Based on the above issues, this thesis presents the general framework and method of ontology learning based on granular computing in the thinking way of human learning. According to the task of ontology learning from bottom to top level, there are the concept learning, taxonomy learning, term learning, non-taxonomic relation and rule learning in the general framework. Meanwhile, the ontology concept and taxonomy are extracted from the sets of domain instances in the ontology learning method, so as to construct the ontology concept granular space. The data of extracting from heterogeneous data sources is preprocessed to the information input system for ontology learning, therefore, the generality of ontology learning has been greatly improved. The proposed Ontology learning method is applied into the process of person domain ontology learning, so as to get the person domain ontology. Furthermore, this thesis combines the person domain ontology and search engine together, and presents a novel search engine which not only improves the accuracy of person information retrieval, but also verifies the correctness of person ontology and effectiveness of the ontology learning method.· Through the analysis of the ontology learning research situation and the existing problems, this thesis presents the general framework of ontology learning. Firstly, from the perspective of ontology organization structure and heterogeneous data sources, this thesis describes the definition and classification of ontology, ontology learning and isomorphic ontology learning information input system. Secondly, according to the characteristics of the data in the ontology learning information input system and the thinking way of human learning, this thesis presents the general framework of ontology learning. On the basis of the task level of the general ontology learning framework, this thesis proposes the corresponding ontology learning method. It provides the basis for subsequent research on ontology learning.· According to the needs of ontology learning in no initial ontology situation, this thesis presents ontology concept granular space model and the algorithm of construction ontology conceptual granular space. Firstly, this thesis describes the rough set representation and granulation method of domain instance and concept grain, the relationship and arithmetic between ontology conceptual grains, the rough similarity representation method of domain instance grains. Secondly, according to the different features between domain instance grains and ontology conceptual grains, this thesis presents the algorithms of construction initial abstract conceptual granular space and multi-level conceptual granular space. Experimental results show the effectiveness of the proposed approach, which can outperform the algorithm C5.0 and Kmean.·According to the needs of ontology learning on the basis of initial ontology, this thesis presents the improved algorithm of semantic similarity based on HowNet. Through the analysis the domestic and foreign related conceptual semantic algorithm, this thesis summarizes the mainly factors affecting the concept semantic similarity algorithm, combines with the content organizational structure characteristics of HowNet and presents the improved algorithm that based on the existing semantic similarity algorithm of HowNet. Comparing with the other similarity algorithm, the proposed method can improve the accuracy of conceptual similarity.· The proposed ontology learning methods are applied into the ontology learning process of person domain, and the person domain ontology can be constructed. In order to reduce the time complexity of conceptual granular space and the dimension of attributes in information input system, this thesis presents multi-classes classification algorithm KNN-DAG-SVMs based on the vector space model. Experimental results show the effectiveness of the proposed approach dramatically improves the precision the recall, and outperforms KNN and SVM.· The person domain ontology is applied into search domain. This thesis proposes the framework of smart search based on person domain ontology, and describes the function of all modules in the framework. In order to improve the speed of retrieval results, the three levels inverted index model is presented in the data index module. Meanwhile, the learning algorithm of new domain instance is presented in the ontology learning module and the smart search system based on person domain ontology is implemented. The system not only improves the precision and recall of person information retrieval results, but also further enriches the instances of person domain ontology.