| With the advent of the Internet, there are millions of websites that publish information. Before information from the diverse information sources can be fully utilized, their heterogeneity must be resolved. Although significant progress has been made to resolve heterogeneity due to differences in operating systems, computer architectures, and networking protocols, little progress has been made to resolve the semantic heterogeneity among information sources. This work addresses the problem of semantic heterogeneity and has resulted in the ONION system, a framework proposed to enable interoperation among information sources.; First, the problem of semantic heterogeneity is addressed by showing that semi-automated tools can match ontologies associated with information sources. The ontologies contain specifications of the semantics of the vocabulary used in information sources. SKAT, an articulation rule generation tool, automatically creates a set of suggested articulation rules that specify the semantic correspondences between concepts across ontologies. S KAT consults dictionaries and thesauri to find word similarities, heuristically determines phrase similarities using a corpus-based information retrieval technique, and also utilizes structural similarities between ontology graphs to determine matching rules. These rules are then verified by a human expert. The responses of the human expert are logged for future use while generating articulation rules between similar ontologies. In our experiments, SKAT generated 75% of the matching rules average on average with less than 10% erroneous rules.; Second, an ontology-composition algebra is proposed to manage related ontologies, using rules generated by SKAT. The algebra has one unary operator, Select, that helps select relevant parts of an ontology. Three binary operators, Intersection, Union, and Difference, allow composition of selected parts of the ontologies and creation of new ontologies. The properties of the algebraic operations have been characterized and form the basis for optimizing a composition task.; Logging of the expert responses, along with the declarative specification of the composed ontologies significantly simplifies the maintenance tasks that arise when source ontologies are updated. Now, an automated tool can easily replay the composition task to recreate valid composed ontologies after the source ontologies or the articulation rules between source ontologies are changed. |
