Web services composition: a story of models, automata, and logics

A key research challenge in Web services concerns (semi-) automatic discovery and composition of Web services, in order to construct new Web services with desired properties or capabilities. This talk provides a survey of key developments that work towards this ambitious goal. The fundamental work in this area has centered on three models, each coming with a different approach to the composition problem. The OWL-S model for Web services focuses on how Web services interact with the "real world," represented abstractly using (time-varying) first-order logic predicates and terms. A representative composition result here uses a translation into Petri nets. The "Roman" model for services focuses on an abstract notion of "activities" (without modeling how they impact the world), and in essence model Web services as finite state automata with transitions labeled by these activities. A powerful composition result is obtained using a reduction to propositional dynamic logic (PDL). The conversation model focuses on messages passed between Web services, and again uses finite state automata to model the internal processing of a service, with transitions labeled by message sends and message reads. A key result here concerns determination of the "local" behavior of individual services, if they are required to conform to a given "global" behavior. The talk also discusses two ongoing efforts to unify the three models just described. One activity, by the semantic Web services intiative (SWSI), is to develop a semantic Web services ontology (SWSO). This is based on the Process Specification Language (PSL), a first-order ontology for sharing descriptions of manufacturing processes, which recently became an ISO standard.