Attribute-Based Semantic Reconciliation of Multiple Data Sources

Abstract

The main challenge in information integration is reconciling data semantics. Semantic reconciliation usually follows one of two approaches. Schema-based approaches assume instances belong to types and use schema information to reconcile types before reconciling attributes and relationships of these entity types. Attribute-based approaches use attribute names and nom-semantic information, such as ranges of attribute values, to seek similarities in attributes across sources. We suggest an attribute-based approach that uses semantic information. Two principles guide our method. First, reconciliation does not require that instances belong to specific types. Instead, sources can be reconciled by analyzing properties. Second, properties that appear different may be manifestations of the same higher-level property, enabling the treatment of inter-sources attributes, with different names and value ranges, as specializations of a more gemeral attribute. We discuss the approach, analyze its potential advantages, suggest a formalization, demonstrate its use with examples, and suggest directions for further research.