Improved Classification Of Ontology Instance Based On SVM Algorithm

The continuous development of Information Technology and the Internet，not only brought theproblem of“Information Overlap”，but also brought the problems of“InformationScarcity”,”Information Trap”,“Information Fog”and other issues. All of these make it becomingmore and more difficult to obtain the specific information. Some scholars have put forward the ideaof Knowledge Engineering to solve these problems. Knowledge Engineering includes three basictopics, such as Knowledge Representation, Knowledge Acquisition and Knowledge Management.Among of the three topics, Knowledge Representation is the core of Knowledge Engineering. Thepast study has found that ontology is a good representation method to represent domain knowledge.Ontology auto-population is a hot issue of ontology research, it means to automatically processthe corpus to extract the instances, establish relationships between the concepts and instances, andfinally establish a complete and consistent ontology. Ontology instance classification is the key partof ontology auto-population, it means computing the similarity between concepts through specificclassification algorithm. On the one hand, it needs to identify the ontology instance and then to fillthem to the appropriate ontology concept; on the other hand, it needs to identify the ontologyinstance as much as possible.At present, the research of ontology instance classification is still in an immature stage. Mostscholars and institutions do such research by drawing the methods of Artificial Intelligence andMachine Learning. It still has a lot of problems to solve in spite of some achievements. There aretwo main methods to cope with the issues of ontology instance classification. One is the rule-basedmethod, which means to implement the issue of ontology auto-population by constructing rules ofinstance identification and classification. The other is statistical model-based method, which meansto build statistical models by training labeled corpus or self-learning approach.As one of the widely used algorithm in artificial intelligence and machine learning, SVMalgorithm can be used to solve the issue of ontology instance classification. SVM is a vector spacemodel, which simplifies the text processing to the vector operation in vector space. It achieves thegoal of ontology instance classification by calculating the similarity between vectors.After comprehensively analyzing the previous researches, this paper focuses on improving theSVM algorithm. In this paper, the Onto-Bt-SVM model is built to solve the classification issue ofontology auto-population. The strategy to improve the SVM algorithm mainly reflected in two aspects: First, this paper constructs multi-class SVM classifier by taking advantage of the ontologyconcept structures and the binary tree models (BT), in order to making the SVM algorithm beingable to cope with instance classification issue of ontology auto-population. Second, this papermeans to make full use of ontology semantic knowledge to construct feature vector, then choosingmore proper features as the dimensions of feature vector. Generally speaking, the improved modelcan better handle classification issue even under limited sample data. This paper designs fourexperimental programs to evaluate the effect of Onto-Bt-SVM model. The experimental resultsshow that Onto-Bt-SVM model can greatly improve the classification accuracy and recall rate. Inaddition, this paper has also studied the ontology construction method, and then this paper submits acomplete set of ontology construction methodology. Finally, this paper uses the methodology as aguideline to build a new-concept-weapon ontology.