Research On Machine Learning Based On Ontology

Data structure of internet using now, is for users reading directly, and is devoid of semantic information that computer can operate. Therefore, it restricts computer's capability of automatizing and intelligentizing in the domain of information searching. Hereby semantic web and interrelated concept were brought forward and obtained numerous scholars' attention. Ontology is a network including notions, properties and relations, ontology is the base of semantic web. Usually ontology is established by the philologist and the professor of domains. Protege is most frequency tool of ontology building empoldered by the Stanford University, it makes built ontology easier and the ontology built by protege is bases syntax of OWL.OWL is a sort of semantic web language constituted and developed by W3C, OWL is standard ontology describe language at present.At present, some scholars have built their ontology in their project, but ontology required by intending semantic searching are cosmical and relate to a large number of domains. It is an important that how to conformity these demain ontology. The research ontology mapping is to result the problem. Now there are some productions with some methods have been contrived, but they have some limitation and performance isn't satisfied.The concepts, types, method of ontology and a model to ameliorate sorting of ontology mapping will be introduced in this paper. First, the quantity of factors of the ontology mapping will be calculated and the result will be input of BP Neural network in this model, and then searching representative ontology to train the neural network, and modifying the weight from back to front if result of network calculating is differ from the result excepted.This thesis will introduce basic method of neural network designing. An example that use to incorporate multi-strategies based neural network will be illustrated. Including topology structure designing a C language program about neural network, process of training. This system used some optimizing method of weight initializing and alterable study speed to decreasing systemic study time. In test it used ten set of different training examples to test study speed of study arithmetic. The result of test indicates that the two methods make study speed expedite sixty times and settle problem of study speed is tardiness. The neural network of this system can be extended to put the axe in the helve of conformation of complex function.