Module Network Model for Ontology Merging

Ontologies are an important feature of Semantic Web. The massive information created by the exponential growth of webpages requires sharing of ontologies. One of the key features to improve Semantic Web is to develop ontologies. Manual annotations of the webpages require massive amounts of resources. A feasible solution to reduce cost is to build ontologies, merge ontologies and enrich ontologies with changing world. Hence, the development of tools for ontology engineering would be an efficient way to the success of Semantic Web. In this work we describe a new machine learning approach for ontology merging. A novel module network method to merge ontologies is described. Given two or more ontologies, the module network assigns score to the concepts and the learning algorithm searches for the maximum score for merging. The module network approach is a robust probabilistic method which ensures that the noise in the data is minimized. This approach is better than the other machine learning approaches used namely Bayesian network. This approach is less computational intensive than the Bayesian approach. The representation of the variables in the modules significantly reduces the complexity of the model space and the constraints of the variables. A key feature of module networks is the inherent “merging ” of the modules is a basis of merging of the ontologies. The concepts are considered.to be variables of each modules and are merged to the module which produces the highest score. One of the salient features of using module networks is that it provides an insight to the domain structure because the variables in a module share the same parents and have the same statistical properties. The variables of ontology merger are the concepts. The similarity measure that is use in the module network implementation is Kullback-Leiber measure for the initialization of the module network model. We will employ n-fold cross validation to ensure that the model does not overfit the data. A statistical method to measure the success of ontology merging is described in this work. We test the quality of the merging of the ontology with other ontology merging algorithms using Bonferroni correction