Property-Based Semantic Reconciliation of Heterogeneous Information Sources

Integrating information from diverse sources is of great importance in the database area. The main difficulty in information integration is reconciling data semantics. Common approaches to semantic reconciliation are based on first identifying similar entity types in various sources, and then reconciling entity type properties (attributes and relationships). Such approaches assume all instances to be reconciled belong to well-defined types. We suggest an alternative approach based on two fundamental principles. First, reconciliation does not require that instances be assigned to specific types. Instead, sources can be reconciled by analyzing similarities of properties. Second, properties that appear different may be manifestations of a higher-level property that has the same meaning across sources. We present the fundamental ideas underlying our approach, analyze its potential advantages, suggest how the approach can be formalized, demonstrate with examples the feasibility of using it for semantic reconciliation, and suggest directions for further research.