Ontology-Based Integration for Relational Data

Framework and Results. In this paper, we briefly describe an ontology-based approach that uses a first order theorem-prover for information integration. We have also built OntoGrate to evaluate our approach and describe its architecture here. Throughout, we refer to the term ontology as the formal specifications of the vocabularies of concepts and the relationships among them. Also, for us, integration has two aspects: (i) query answering and (ii) data translation. Mappings between schemas are essential to integration and require an adequate representational language. SQL views have been widely used, but mappings can also be represented with other languages having formal semantics (e.g., Datalog, XQuery). Instead, we choose a more expressive first order ontology language, Web-PDDL, to represent complex mappings between schemas as bridging axioms (first order mapping rules). The advantage in doing so is the specialized theorem prover, OntoEngine [2], can then perform query answering and data translation while formally preserving semantics. We refer to this process as inferential data integration because it uses sound inference by either forward chaining or backward chaining. Two databases in the online sales domain, Stores7 from Informix and Nwind from Microsoft, serve as examples in Figure 1. First, we define a super ontology for SQL to express concepts such as aggregate functions and integrity constraints that exploit desirable features of database systems via ontology inheritance. Then, by a simple process, we translate each database schema into its own ontology. Next, we define mappings between each ontology using bridging axioms and call this our merged ontology. Finally, syntax translators that we developed allow OntoEngine to access actual relational data by transforming atomic queries from Web-PDDL to SQL. These elements are summarized in the architecture of OntoGrate shown in Figure 1. Therefore, the user can submit a query or translation request which OntoEngine fulfills by either backward or forward chaining on the bridging axioms in the merged ontology using actual data retrieved by the SQL syntax translators. Although our Web-PDDL-to-SQL translators only handle atomic queries for now, the inference mechanisms can still process more complex conjunctive