The World Wide Web (WWW) contains a vast amount of information valuable for planning and design tasks in engineering. Electronic product catalogues and online repositories, for instance, provide component descriptions which can support the human engineer as well as semi-automatic synthesis programs in finding appropriate parts. Engineering design involves usage of domain specific technical knowledge together with creative problem solving skills to select properly functioning components that comply with a set of requirements like performance goals, physical or cost constraints. Intelligent agent perform specific tasks on the behalf of users. The KRAFT project [PHG99] analyses, how to fuse and transform selected semi-structured Web data and meta-knowledge, especially constraints, into a form which can be applied by intelligent agents in order to support further knowledge processing.
Web resources related to a domain may differ in syntax, structure and semantic content. 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 information requests can be transformed, distributed and matched by information offers which are stored as concrete resources. Common interchange formats, document type definitions (DTDs), and shared meta-level knowledge and data, like standardised conceptual models, called ontologies, have been identified as central issues for the construction of distributed information systems which take their input from Web resources. The eXtensible Markup Language (XML) has emerged as a new standard for exchanging and representing data and structure on the Web. For a wide range of applications, XML allows user to define the structure of documents by means of Document Type Definitions. A DTD is essentially a formal grammar which restricts the structure of valid documents. Semantically similar product categorisations may coincide or differ in various structural aspects, such as equivalence, overlap, and inheritance of attributes or arithmetic measurement of values. An ontology is a formal explicit specification of an agreed standardised vocabulary, a precise definition of the basic terms and the relationships that can exist between them [Gru93]. Hence, the use of terms from an agreed ontology in product descriptions is suggested as means to circumvent misunderstandings and support mediation. However, in many cases, the mediation task may require not only to match structures and semantic contents of documents, but more elaborate reasoning as well.
Among the many data models that can be used to enable reasoning and construction with the semi-structured information contained in Web resources and associated meta-information and knowledge, it is believed that graphs are the most adequate ones. Graphs provide natural algebraic means, for example projections and equivalence, to model knowledge transformation and sharing. Most important, labelled directed acyclic graphs offer a formal set-theoretic data model for subsets of the standard XML for exchanging Web data and meta-data, and for the design of query languages as XML-QL (cf. [DFF98]). Interacting constraints and constraints which apply to the structure and content of Web resources are essential for planning and synthesis tasks.
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. However, it is a standard practice to combine logical reasoning and constraint solving in ``constraint logics''. In order to combine the advantages of an efficient graph-based handling of knowledge representation with the expressiveness and reasoning facilities of constraint logic, this paper analyses a logic-based representation framework which allows for efficient graph operations without loosing the reasoning and constraint solving capabilities of constraint logic. It is based on graph-based data models for XML (cf. [GMW99]) and exploits a convenient formal correspondence between mathematical semantic structures (DAGs), syntactic term encoding, and logical specifications. This correspondence is extensively exploited in the SEAMLESS framework (cf. [Eus00]) which allows to specify and solve complex synthesis problems as is required for the mediator tasks described above. A graph term algebra and canonical forms for classes of equivalent objects which enable efficient graph operations within a logic-based framework are employed to implement the functionality.
The remainder of this paper is organised as follows. Section presents an architecture for knowledge mediation on the Web. Section describes the graph-based knowledge representation language. Section shows how to formalise ontologies as concepts with attached constraints. Section sketches the synthesis of rules that link ontological descriptions and DTDs, and describes the mediator shell. Finally, a summary concludes this paper.