Using Lexical And Semantic Analysis For Ontology Integration

An ontology, as a formal explicit specification of a shared conceptualization, is utilized to share, reuse and interoperate knowledge between soft agents. Currently, it has been adopted widely in many applicaitons such as semantic web, knowledge and data engineering, and semantic IoT (Internet of Things). However, most of these ontologies in similar (or same) domains are constructed and maintained by different knowledge engineers who have various backgrounds and use different terminologies to describe concepts. The heterogeneity between different ontologies for representing similar (or same) domains blocks sharing, reusing and interoperability across ontologies in application systems. Ontology integration methods deal with this issue. Its main task is to establish correspondences between entities, each from heterogeneous ontologies respectively derived from different data sources which represent similar (or same) domains, and then construct an integration ontology with the established correspondences. This dissertation discusses some key issues in an ontology integration, which include lexical analysis, semantic analysis, the construction of a consistent integration ontology and so on.The main contributions of this dissertation are listed as follows:1. Analyzing lexical information based on an entity notion. Considering special lexical relations in WordNet, this dissertation extends the sense of each word to a set of senses. This helps to include the possible senses of the word in some existing context instead of only finding the suitable one and supports potential relations between entities. This dissertation also formally defines an entity notion combining syntaxes (such as(?),(?) and3) with senses of words. By analyzing evaluations, this method descreases the precision of the system by3%, but helps to increases the recall of the system by37%.2. Proposing an iterative and semantic filtering process to remove redundant matching candidates. We firstly draw up rules for filtering; secondly present a filtering algorithm; finally formally define sufficiency conditions for filtering and prove that the process is terminable. By analyzing evaluations, this method enhances the accuracy of the system. 3. Formally defining the closest subsumer of an entity description by thinking the least common subsumer and the most specific concept in description logics. Based on the definition, we produce the supplementary correspondences based on the filtered matching candidates, which exist in standard alignments but seldom appear in the alignments of other solutions. By analyzing evaluations, the definition helps to improve the recall and precision of the system.4. Extending structural subsumption reasoning algorithms to analyze semantic information. This dissertation firstly parses various constructors and axioms in an ontology to make the implied semantics and lexical information to be combined together and read-off easily by rephrasing entities into normal forms based on the representation of an entity notion; and then compare the syntactical structures of normal forms, by zooming in (out) the sets related to the entities and adjusting the allowable degree for discrepancy between entities, to infer corrrespondences. By analyzing evaluations, the method for analyzing semantic information increases the recall and precision of the system.5. Formally defining an ontology integration based on micro-sub graphes and proposing algorithms for constructing an integration ontology. The algorithms donot break semantic structure in heterogeneous ontologies and promise that the integration ontology adopte the most correspondences. The correspondences only bridge the heterogeneous ontologies. We also prove the established integration ontology is consistent.