The WWW can be considered as market place where services and products are freely offered and requested. The general idea of a mediator system is to integrate applications and heterogeneous information sources in such a way that they keep their autonomy (cf.\ [Wie92]). A mediator has to ensure that semantic knowledge requests from users or applications can be transformed, distributed and matched by information offers which are stored as concrete resources, having resolved semantic ambiguities. Matching service descriptions have to be transformed back into high-level information, suitable for further processing. The KRAFT project has developed an agent-based architecture for distributed query processing [PHG99]. All information processing units are realised as interacting software agents, using a subset of the Knowledge Query and Manipulation Language (KQML) performatives [FYM97]. The facilitator maintains a data base (``yellow pages'') of all registered service providers and is able to map service types to appropriate providers. A provider first has to export the type of service it wishes to advertise via the facilitator. Information resources are coupled to the mediator, using wrapper modules. Information is exchanged in the KRAFT Constraint Interchange Format (CIF).
Figure: An Architecture for Mediation
With the vast amount of information and hidden knowledge in Web
resources, standards as XML and DTDs for structuring, typing and
semantic conceptual abstractions as shared ontologies have become
evident. Using current data models for WWW information, there seems to
be no way of declaring basic types and to constrain admissible values
in the same formalism. The presented work establishes means that
enable integration, representation and reasoning using Web standards,
semantic concepts, and domain-specific knowledge like
constraints. Mediation tasks as query processing and integration of
heterogeneous information sources can be handled more
effectively. Problem solving tools can re-use integrated and
transformed knowledge for the automated solution of complex
tasks. Knowledge integration and transformation is achieved by
adopting the SEAMLESS multi-layer architecture (cf.\
[Eus00]). DTDs and ontology are incorporated into the
distributed architecture through an intermediate context-dependent
mediator knowledge base that in SEAMLESS terms provides a
viewpoint. Figure 
demonstrates the resulting
architecture. The mediator architecture thus employs three essential
technologies which will be shortly examined: "XML", "Shared
Ontology" and "Viewpoints".
