Abstract

Ontologies are increasingly being recognized as a critical component in making networked knowledge accessible. Software architectures which can assemble knowledge from networked sources coherently according to the requirements of a particular task or perspective will be at a premium in the next generation of web services. We argue that the ability to generate task-relevant ontologies efficiently and relate them to web resources will be essential for creating a machine-inferencable “semantic web”. The Internet-based multi-agent problem solving (IMPS) architecture described here is designed to facilitate the retrieval, restructuring, integration and formalization of task-relevant ontological knowledge from the web. There are rich structured and semi-structured sources of knowledge available on the web that present implicit or explicit ontologies of domains. Knowledge-level models of tasks have an important role to play in extracting and structuring useful focused problem-solving knowledge from these web sources. IMPS uses a multi-agent architecture to combine these models with a selection of web knowledge extraction heuristics to provide clean syntactic integration of ontological knowledge from diverse sources and support a range of ontology merging operations at the semantic level. Whilst our specific aim is to enable on-line knowledge acquisition from web sources to support knowledge-based problem solving by a community of software agents encapsulating problem-sloving inferences, the techniques described here can be applied to more general task-based integration of knowledge from diverse web sources, and the provision of services such as the critical comparison, fusion, maintenance and update of both formal informal ontologies.

Author Keywords: ontological knowledge; software architecture; domain knowledge; semantic web; information integration; task models.

^f1 lrc@psychology.nottingham.ac.uk.

^f2 nrs@ecs.soton.ac.uk.