The Service Classifier Agent


UIAs, QPAs and Auction agents use the same SCA, and thus subscribe to the same ontologies: the UMDL SMS is not terminologically heterogenous. Figure 3 illustrates the SCA's interactions with the other agents. Agents are in ovals, ontologies in rectangles, and the arrows connecting agents represent messages, paraphrased in natural language. In the SMS these messages are expressed in Loom. To communicate with the SCA, agents use terminology from the nested ontologies. The thin line around the SCA is jagged to show that the set of available terms is dynamic; both QPAs and the AMA add concepts to the agent services ontology. Messages use font styles that correspond to the ontology labels (plain, italics, and upper-case), to highlight the source of their terminology.

Figure 3: Agent interaction with the SCA in the SMS

Figure 3 illustrates the ability of the SCA to construct new terminology from existing terms at runtime. Concepts in the SCA's dynamic ontology hide knowledge, just as words chunk meaning in natural language. Auctions sell some service, but they don't need to know anything about that service. The AMA asks the SCA for a service label for an auction, using the QPA's service label, but it does not know anything about that service. The SCA classifies the auction service using characteristics inferred from the QPA's service label. Thus, the SCA can respond to the UIA's request for an auction, which is phrased in terms from the static ontologies, rather than the label which the AMA used to define the auction service. This appropriate hiding of knowledge reduces overall system complexity, and increases reusability and maintainability.

In future versions, auction managers will make extensive use of the SCA's knowledge. First, ontologies are the natural place to represent the type of auction that is appropriate for marketing a particular service. Second, auction managers may need to define markets for sets of services, rather than services with exactly the same classification. Efficient resource allocation depends on maintaining suitable numbers and ratios of seller and buyer agents. These quantities can be controlled by adjusting the generality or specificity of the service to be auctioned.

The most dramatic contribution of the SCA to the SMS derives from its declarative description of services, and its ability to rank available services given a target and search strategy. When a new QPA provides a service that better meets a UIA's needs, the UIA switches to buying services from the new QPA, without requiring any intervention to modify the UIAs behavior. This occurs because the UIA periodically repeats its search for the best available service. Furthermore, if the new service subsequently becomes unavailable, the UIA reverts to its previous supplier.

Jose M. Vidal
Tue Sep 30 14:35:40 EDT 1997