An Extensible Model-Based Mediator System with Domain Maps

Mediator systems federate and integrate data from disparate sources in order to elicit information that the individual sources cannot provide independently. The standard mediator architecture employs wrappers that translate heterogenous source data into a common (often semistructured) data model like XML. A “mediation engineer” provides an integrated view definition (IVD) on the wrapped XML sources. In such a system, an IVD is ideally expressed in a declarative query language for XML or semistructured data. When developing the IVD, an XML query language provides the mediation engineer only with a tree-structured model of the source, i.e., the names and possible nesting structure of XML elements as defined by an XML DTD, but gives no hint on semantic relationships, class structures, not to mention application domain specific constraints. Indeed, as shown in [4, 1, 3], mediation should be lifted to the conceptual level when mediating across complex sources whose data comes from seemingly disjoint “worlds”, e.g., two neuroscience labs creating information on neurotransmitters and protein distributions, respectively.1 To this end, we present a mediator prototype system whose main novel features are: (i) mediated views are defined and executed at the level of conceptual models (CMs) rather than at the usual structural level, (ii) domain maps (DMs) – labeled graphs of concepts and relationships with a formal logic semantics – are used to bridge the semantic gap between source data from “multiple worlds”, and (iii) a plug-in mechanism for CMs and DMs is provided which allows the mediator system to be easily extended when new formalisms for CMs and DMs are used by sources. We illustrate these features using an example from a complex neuroscience mediation problem. For details of model-based mediation with domain maps, including their formal semantics, see [4].