An Ontology Server for Agentcities.NET Project

The aim of this project is to investigate the use of an ontology server (or registry) in an interoperable agent network (agentcities). The registry provides a publication environment for the disclosure of metadata vocabularies and customised application-specific profiles of these vocabularies. The metadata vocabularies (also known as schemas or metadata element sets) may be regarded as simple forms of ontologies. In the registry environment, individual terms as well as whole vocabularies can be investigated for adaptations, local usages and relationships with other vocabularies. We are building on previous work within the SCHEMAS and MEG projects, and on our involvement in the Dublin Core Metadata Initiative's registry activity.

The project started on 1^st September, 2002.

We have installed JADE and set up the server, which will host the ontology service, and we have registered our platform with the platform directory at www.agentcities.net. Our platform name is ukoln.agentcities.net.

The research team in the Department of Computer Science, at the University of Bath, led by Dr. Julian Padget, have been providing support in setting up our Agent platform. We will be liaising further with our colleagues here in Bath whilst investigating the behaviour of the ontology server. In particular, the ontologies being developed for describing accommodation in the tourism sector, in the Bed and Breakfast Reservation Service project, may provide a realistic data set for the ontology server.

As a point of departure, we are currently evaluating the registry software developed in the MEG project (which in turn drew on the work of the DESIRE and SCHEMAS projects). All of these projects explored approaches of declaring and sharing metadata vocabularies using RDF Schema. The existing registry software stores information pertaining to metadata vocabularies (or Elements Sets) and provides an interface for interacting with the information.

The registry is based on the following model of metadata vocabularies:

Element Sets are owned and maintained by Agencies. Element Sets are made up of Elements. An

Element Usage may:

• introduce constraints on the value of an Element by associating it with one or more Encoding Schemes;
• introduce constraints on the obligation to use an Element (e.g. make its use mandatory) or the occurrence of an Element (e.g. whether it is repeatable);
• refine the semantic definition of an Element to make it narrower or more specific to the application domain.

Encoding Schemes constrain the value space of Elements. An Application Profile defines a set of Element Usages of Elements drawn from one or more Element Sets.

• Element Sets (i.e. on the Element Sets as units, rather than on their constituent Elements), including information on their intended scope/area of use and their relationship to other Element Sets;
• the Elements which make up those Element Sets, including information on the semantics of the Elements and their recommended usage, and any semantic relationships to other Elements in this or other vocabularies (e.g. the relationship described by the DCMI concept of "element refinement" or by RDF Schema as a "sub-property" relation)
• Application Profiles, including information on their intended scope/area of use and their relationship to other Element Sets and Application Profiles;
• the Usages of Elements which make up those Application Profiles, including the Element used, any prescription of Encoding Schemes, and other constraints on element use;
• Encoding Schemes, which constrain the value space of Elements, including information on their intended scope/area of use; where an Encoding Scheme takes the form of an enumerated list, the values prescribed by that Encoding Scheme may be recorded;
• the Agencies who own/create/maintain Element Sets, Application Profiles, and Encoding Schemes

The information is stored and made available in machine-processible format as schemas. The W3C Resource Description Format is used, and access to the registry utilises the RDF toolkit, Redland. The existing registry API is developed in Perl and supports functions such as querying of the registry.

A simple deployment of the registry within an Agent environment can be achieved by developing and deploying a Registry Agent on the UKOLN agentcities platform. The Registry Agent will handle queries from other Agents wishing to interact with the registry, using the specified Agent protocols, interacting with the registry through the Registry API.

Using the above deployment model, the behaviour of the Registry Agent will depend to an extent on the capabilities of the Registry API, for example the complexity of the searches that may be carried out depends on the granularity that the registry API affords. One research issue is whether the existing API supports sufficient functionality to satisfy the required Agent behaviour. We will investigate possible models of agent behaviour relative to registry capabilities. This will be informed by our internal expertise, and through liaison with the Computer Science staff at Bath.

Commitment to an ontology exchange format (and vocabulary description language) requires further study. There are various proposals in this area, as reviewed in the OntoWeb Technical Roadmap. Some are the result of various W3C groups, other proposals may come from the agent community itself. The Agentcities Ontology working group suggests a number of tools that may feed into the study of the adoption of a specific framework, all of which can handle RDF Schemas. UKOLN is currently contributing expertise in RDF Schemas as a means of ontology exchange; RDF forms a foundation of many of the extensions being considered (such as DAML+OIL and OWL)

Clearly the registry agent as depicted in Figure 1 is required to interact with other agents, and the underlying registry. Research is still required to understand precisely what requests (expressed in ACL) from other agents the registry must handle. Work is also required to assess models of response. For example, should the registry agent respond to requests by simply wrapping data retrieved from the registry within a content response parameter, or can/should the response be more structured, possibly via additional dialogue between the agents?

This work will provide valuable feedback and experience on the strengths and weaknesses of the current approach to registry behaviour, whilst the outcomes will feed into, and help to shape other experimental architectures and models of ontology exchange. Ontology servers are an essential part of the infrastructure required to enable the exchange and re-use of ontologies. The provision of semantics in a machine-processible format will be a key achievement in the road to semantic interoperability.

[Note: this is the latest version of the W3C Working Draft, released on 12 November 2002, which is a work in progress; The registry development took place before the release of this draft.]

DAML+OIL http://www.w3.org/TR/daml+oil-reference

OWL http://www.w3.org/TR/owl-ref/