Knowledge discovery from databases on the semantic Web

We provide a flexible method for knowledge discovery from semantically heterogeneous data, based on the specification of ontology mappings from the local data sources to pre-existing (superior) ontologies in an ontology server. We also provide an innovative method for the construction of a dynamic shared ontology; data integration is then carried out by minimisation of the Kullback-Leibler information divergence using the EM algorithm. The new knowledge learned by this process is potentially richer than any of the contributing data sources. We also show how the approach may be extended to knowledge discovery from a number of database attributes; association rules or Bayesian belief networks may then be induced. An architecture for a KDD system in such an environment is described; this is an extension of a previous architecture for distributed data processing that we have already implemented.