Upload
everett-chapman
View
215
Download
2
Embed Size (px)
Citation preview
Knowledge Mediation in the WWW based on Labelled DAGs
with Attached Constraints
Jutta Eusterbrock
WebTechnology GmbH
Introduction
• WWW: Vast amount of useful data and information in online repositories, electronic product catalogues, ... for configuration, planning, synthesis applications
• Problems: – WWW topology is dynamic, content changes quickly– Information from various locations differs in syntax,
structure, semantics– Domain-specific meta-knowledge and interacting
constraints need to be taken into account
Introduction
• Goal: Use of WWW data from various locations, attached meta-knowledge and constraints for applications, eg. configuration, based on reasoning and constraint-solving
• Approach– Agent Framework– Viewpoint as Mediator (Intermediate layer between
resources and applications)– Logical Representations of XML-Graphs for Data
Integration
Knowledge Integration XML DTD
• XML (eXtensible Mark-up Language)– Emerging standard for exchanging data on the WWW
– Objects consist of nested elements,attribute/value pairs
• DTD (Document Type Descriptor)– Grammar, Vocabulary (optional)
Knowledge Integration XML Query Language
• Access to fragments of XML elements through a number of query languages, based on path-expressions – Example in XML-QL [Deutsch, Fernandez, Florescu, Levy,
Suciu, 98]
Knowledge Integration Agents
• Application of the KRAFT Agent Framework– Software components realised as interacting agents– Subset of KQML performatives for communication – Facilitators encompass descriptions of service
providers that have to advertise their capabilities– Wrappers translate and distribute queries– Mediators provide uniform access to heterogeneous
information resources
Knowledge Integration Knowledge-Bases
• Shared Ontology– Formal semantic domain model– Explicit specification of agreed, standardised
vocabulary, definition of the basic terms (concepts), properties, relationships (Gruber) and background knowledge
• Facilitator Knowledge Base– Representations of syntactic Web (meta-) data,
schemas, locations• XML elements, DTDs• Stored as facts in KB
Knowledge Integration Viewpoint
• Realisation of Mediators by Viewpoints – Provide context-specific definitions for the ontological
concepts– Based on lifting rules (articulation axioms, Guha, Cyc)– Interpretation with respect to a semantic requests
generates a set of syntactic queries to individual resources by reasoning, constraint-solving
– Knowledge Integration with respect to a given ontology
Graphs for Web Knowledge Bases
• Wrapping, storing, retrieval, reasoning based on a logical representation of graphs– Labelled DAGs as data model for XML elements
– Feature Graphs for modeling DTDs and concepts
– ADT for labelled DAGs and efficient canonical term encodings (Eusterbrock, 97)
• Graph retrieval based on path-expressions
• Graph matching modulo isomorphism by term matching
• Graph term subsumption models class-, instance relations
Domain Ontology with Constraints
• Representation of domain concepts by feature graphs with attached constraints
Knowledge-Based Mediation: Objective
• Queries can be expressed– using the vocabulary of a shared ontology
– built-up as conjunction of atoms, graph-path expressions and constraints attached to free variables
– without having to take into account location
Knowledge-Based Mediation: Sharing
• Local Domain Models: Facts based on DTDs• Ontology: CLPs with embedded feature graphs
for concepts• Integration: Translation DTD concept• Semantic mismatches still need to be resolved!
– missing, overlapping features
– feature semantic: prices before/after taxes
– domain values: measurement of units, dimensions
• Sharing Rules: AtomicConcept <= Constraints /\ DTD
– Examples: Specialisation, Unit Conversion
Knowledge-Based Mediation: Sharing
Knowledge-Based Mediation: Sharing
• Sharing Rules: AtomicConcept <= Constraints /\ DTD
– AtomicConcept <--> DTD all kinds of graph mappings, e.g. renaming, projection,
• Lifting– Causal relation between a common aggregated concept
and the set of associated local context (DTDs)– ComposedConcept(_,Subconcept1,...,Subconceptn) <=
FusedConstraints /\ DTD1 /\ ... /\ DTDn
– Linearisation of composed concepts
Knowledge-based Mediation:
• (Automatic) synthesis of lifting rules– graph rewriting, constraint fusion, using sharing rules
Knowledge-based Mediation: Method
• Transformation: Selecting concepts that match graph-paths, rewriting query, using ADT graph
• Decomposition of logical query into queries to individual resources and composition of results– Interpreting lifting rules, background knowledge by
reasoning, constraint-solving yields atomic queries
– Locating suitable resources using facilitator KB
– Wrapping, distributing, querying XML resources
– Integrating returned results into CLP. Non-satisfiability causes backtracking.
Discussion (Some Related Work)
• Mediator systems– KOMET,HERMES logic, deduction for mediator
– Focus largely on uniform access to DBs
• Ontology-based semantic access– ONTOBROKER
• Framelogic for encoding of ontologies
• Generation of DTDs directly from ontologies
Discussion (New Results)
• Lifting rules and viewpoints: – Flexible framework for loose integration
• Graph encoding of XML data, DTDs and concepts– Natural models for knowledge sharing
– Construction of wrappers is straightforward
– Canonical term encodings provide efficient procedures
• Constraints as meta-knowledge– Essential for the automatisation of design tasks
– Rewriting, constraint-solving is automated