next up previous
Next: Graphs for Structured Web Up: An Architecture for Web Previous: Shared Axiomatic Ontology

Mediation by Viewpoints

From the user's perspective, an ontology may provide a meta-level description on the underlying data. In this sense, ontologies act as generic semantic types, that is, they are intensional conceptual descriptions. A concrete information resource, for example a concrete document, may implement an abstract type with certain changes in structure, syntax and symbol names. In concrete inference processes about problem-related facts which need to be extracted from Web documents the system has to rely on knowledge about the related DTDs.

Applying the SEAMLESS architecture, a mediator knowledge-base can be constructed as a viewpoint. Generally, a viewpoint is an intermediate, loosely coupled component that provides the assumptions under which meta-level axioms become true in a particular context. It describes how to "lift" facts and data from the consulted local repositories to a more abstract level so that they can be compared and integrated with others from different sources. A viewpoint may contain additional domain knowledge that supports the integration task.

The viewpoint concept is applied to set-up mediator knowledge-bases that combine a shared ontology and DTDs. Lifting rules shall make causal relationship and applicable constraints among DTDs and a given shared ontology explicit. Using viewpoints supports easy and flexible construction of the mediator knowledge-base in the highly dynamic WWW context. On the basis of a shared ontology, users or software agents can pose semantic queries to the mediator knowledge-base without having to consider the location, the format or context-specific information of the underlying data. The mediator uses knowledge-based reasoning and embeds constraint solving to generate precise data requests. Individual records will then be retrieved using wrappers. Wrappers transform queries from the logical representation format to the existing XML source format, for example XML-QL, using the parsing rules and facilitator knowledge about locations and distribute them in a bi-directional way.


next up previous
Next: Graphs for Structured Web Up: An Architecture for Web Previous: Shared Axiomatic Ontology

Andre Everts
Mon Sep 4 20:44:27 MET DST 2000