Queries, the Missing Link in Automatic Data Integration

This paper introduces the ontology mapping approach of a system that automatically integrates data sources into an ontology-based data integration system (OBDI). In addition to the target and source ontologies, the mapping algorithm requires a SPARQL query to determine the ontology mapping. Further, the mapping algorithm is dynamic: running each time a query is processed and producing only a partial mapping sufficient to reformulate the query. This approach enables the mapping algorithm to exploit query semantics to correctly choose among ontology mappings that are indistinguishable when only the ontologies are considered. Also, the mapping associates paths with paths, instead of entities with entities. This approach simplifies query reformulation. The system achieves favorable results when compared to the algorithms developed for Clio, the best automated relational data integration system. We have developed an Ontology-based Data Integration (OBDI) system that departs from the conventional OBDI organization. The goal is to include automatic integration of new data sources, provided those data sources publish a self-describing ontology. A consequence of that goal is there is no longer the opportunity for an engineer to review and correct an ontology matching prior to its use by the query reformulation system. As ontology matching is understood to be an uncertain process, some other method of mapping refinement is needed. Our system uses queries for this purpose. Ontology mapping in conventional OBDI systems is determined prior to, and without knowledge of the queries to be executed [3]. A static representation of a mapping between target and source ontologies serves as input to a query reformulation module (Fig. 1(a)). In the system described here, ontology mapping is a dynamically computed component whose result depends on the query that is being processed (Fig. 1(b)). In effect, the query becomes a third argument to the ontology mapping algorithm. The query provides context for selecting among competing mappings. Since a mapping is specific to a query, the results may be limited to the partial mapping required by the query reformulation system. The organization was motivated by the following observations. A mapping method may determine that an entity in one ontology maps with equal likelihood to two or more entities in the other ontology. The mapping and reformulation of certain queries is correct only if one pairing is chosen. The correct choice may be different for different queries. The query itself may lend additional semantics that correctly resolve the ambiguity. These observations are supported by the example in Fig. 2. Looking at the ontologies alone, there is insufficient information to determine if the class T :People should Ontology Mapping Query Reformulation Ontology T Query q Ontology S Reformulated query (a) Traditional Ontology Mapping Query Reformulation Ontology T Query q Ontology S Reformulated query