250 likes | 340 Views
Semantically Federating Multi-Agent Organizations. R. Cenk ERDUR, Oğuz DİKENELLİ, İnanç SEYLAN, Önder GÜRCAN. AEGEANT-S Group, Ege University, Dept. of Computer Eng., Bornova-İzmir, Turkey. AEGEANT-S Group -1/2.
E N D
Semantically Federating Multi-Agent Organizations R. Cenk ERDUR, Oğuz DİKENELLİ, İnanç SEYLAN, Önder GÜRCAN. AEGEANT-S Group, Ege University, Dept. of Computer Eng., Bornova-İzmir, Turkey.
AEGEANT-S Group -1/2 • The “Agent and Semantic Web” (Aegeant-S) group in our department works on several projects related with agent systems. • Our main and the most important project is to develop a semantic web enabled and FIPA-compliant agent platform. => “SEAEGEANT” • “SEAGEANT” is a pioneering work in this area. • The name “SEAEGEANT” is chosen to emphasize that the final product is asemantic webenabled agentplatform and it is developed in Aegean (Ege) University.
AEGEANT-SGroup -2/2 • Staff: • Assoc. Prof. Dr. Oğuz Dikenelli (Coordinator) • Asst. Prof. Dr. R. Cenk Erdur • 4 Ph.D. student research assistants. • M.Sc. and undergraduate students. • Project Web Address: http://aegeants.ege.edu.tr
The Problem Tackled in this Paper • With the advances in the agent development frameworks, tools and methodologies, it now needs less effort and time to develop a multi-agent system for an organization. As a result of this progress, we expect a large number of multi-agent organizations to exist on the Internet. • In such an environment, one of the major challenges will be the establishment of an infrastructure for the co-operation of multi-agent organizations providing services in similar domains.
Co-operations between Multi-agent Organizations If request cannot be satisfied here, then the most semantically related organizations should be found... Other multi-agent organizations providing services in similar domains. For example, in “Tourism” domain. Org.-1 ... ..... ? ? ... ... Org.-2 Org.-3
Proposed Solution –1/3 • For the co-operation of multi-agent organizations, an infrastructure providing the semantic interoperation of the multi-agent organizations has been proposed.
Proposed Solution –2/3 • The concept of “Federation” has been proposed. • A federation defines the common characteristics of multi-agent organizations in a domain. • The federation directory service and the federation ontology service are introduced for discovering the semantically related multi-agent organizations in a federation. A semantic matching algorithm has been used in the federation directory service.
Proposed Solution –3/3 • Currently, there are some FIPA (Foundation for Intelligent Physical Agents) specifications discussing the interoperability of multi-agent systems, but these specifications define the standards for interoperability in message transport protocol level, not in semantic level. • FIPA’s message transport level standards are of course important, but are not enough for high-level co-operation. Hence, in this paper we try to fill in this gap.
The Concepts of Semantic Interoperability Federation -1 Directory Service Federation -1 Ontology Service Federation – M Ontology Service Federation - M Directory Service Federation -1 Federation -M Multi Agent Org. -1 Multi Agent Org. -K Multi Agent Org. -1 Multi Agent Org. -K Multi Agent Org. -1 Multi Agent Org. -K ..... ..... ..... Platform-N Platform-1 Platform-2
Platform • Physical infrastructure in which the agents are deployed. It includes basic services as: • Agent management service • Agent directory service • Message transport service • For example, JADE or RETSINA are implemented agent platforms. FIPA, defines the standards for an agent platform. If a platform implementation follows the FIPA standards, then it is called as a FIPA-compliant platform, such as JADE.
Multi-agent Organization • A multi-agent organization is a multi-agent system designed and implemented based on specified requirements in a specific domain. • A multi-agent organization uses the basic services provided by the platform on which it is deployed. • We try to establish an infrastructure for the co-operation of multi-agent organizations; hence, at the conceptual level each multi-agent platform is taken as a seperate entity in terms of semantic interoperability.
Federation –1/2 • A federation established for a specific domain defines the common characteristics of the multi-agent organizations in that domain. In addition, federational rules such as pricing policy, certification can also be defined. • We use the concept of federation not in the same sense with the federation concept in the federated architecture, which has been defined by (Genesereth and Ketchpel, 1994) as an alternative to direct communication in agent systems.
Federation –2/2 • Multi-agent organizations register to the “Federation Directory Service - FDS” using a specific FDS ontology. • The ontologies in a federation are maintained by the “Federation Ontology Service – FOS”. • FDS, is the key element in the architecture, since it can be searched semantically to discover the semantically related multi-agent organizations.
Conceptual Architecture Federation Multi-agent Organization-1 Platform Multi-agent Organization-N Federation Directory Service (FDS) Federation Ontology Service (FOS) ... Agent-K Agent-N : PMS Agent-1 Directory Service Matching Engine ACL Msgs Agent Management Service Message Transport Services of Other Platforms Over IIOP Protocol Message Transport Service
Ontological Infrastructure • There are three kinds of ontologies in the proposed architecture: • Organization Directory Service Ontology • Federation Directory Service Ontology • Domain Ontology
Organization Directory Service Ontology • This ontology is used by the agents in a multi-agent organization for registering their services to the organization’s directory service. The requester agents in a multi-agent organization also use this ontology to search their local organization’s directory service. We can say that this ontology is the internal ontology of a multi-agent organization in a specific domain.
Federation Directory Service Ontology • A multi-agent organization uses this ontology to register its meta-knowledge to the federation’s directory service. • This ontology is also used in the searching of the federation directory service in order to match the semantically related multi-agent organizations.
Domain Ontology • A domain ontology consists of domain concepts, relations between these concepts and rules. • Both the organization directory service and the federation directory service ontologies have a domain specific part, which is expressed using terms from the domain’s ontology.
Comparing the Domain Specific Parts of the FDS and Organization Directory Service Ontologies • The domain specific part of the federation directory service (FDS) ontology is less detailed and contains more high level concepts than the organization directory service ontology. This is because FDS ontology aims at classifying the organizations at a higher level, while the organization directory service ontology is used in representing the services of individual agents within an organization.
General Structure of the FDS Ontology Multi-agent Organization Description Domain Domain_name Domain_ontology_name Domain_dependent_ advertisement_content MTS_Address of the Platform Supported_Content_Languages Supported_Ontology_Languages Supported_Protocols Supported_Encoding hasDomain This attribute is expressed in a language such as OWL and constitues the domain dependent part of the FDS-Ontology. hasProvider Provider // Attributes related with the provider organization such as contact address, web-site, etc.
Collaboration Diagram for Semantic Matching of Multi-Agent Organizations Local Directory Service 1 2 3 FDS Remote Directory Service 4 PMS 5 7 Agent 6 8 Remote Agents 9
Semantic Matching Algorithm in the FDS • It is derived from the previous work on semantic matching of web service capabilities: • (Paolucci, M. et. al., 2002) • (Li, L. and Horrocks, I., 2003)
Semantic Matching Algorithm in the FDS Match(expression, Multi-AgentOrganizationDescriptions) { degreeOfMatch = expression.getOperation(); request = expression.getValue(); matchedOrganizations = emptyList; forall org_descp in Multi-AgentOrganizationDescriptions do { forall adv in org_descp.getAdvertisements() do { if (calculateDegreeOfMatch(request, adv) == degreeOfMatch) then matchedOrganizations.append(org_descp) } return matchedOrganizations; } }
Conclusion and Future Work • Establishing agent federations and using semantic matching to discover a related multi-agent organization in a federation: • Forms the basis for constructing open, large-scale and scalable multi-agent systems. • Makes it possible for multi-agent organizations in similar domain to co-operate effectively. • The idea of confederation based on agent federations will be considered as a future work.
Thank you for listening... Contact: erdur@staff.ege.edu.tr