This is a new working group led by Joerg Hoffmann, concerned with the automization of web service composition.
This component provides support for composing web services into more complex services fulfilling more advanced tasks. To do so, ultimately the developed techniques will have to take all the relevant details into account, ranging from the relevant ontologies and their interrelation to the choreographies of the involved web services. At a more abstract level, functional level descriptions (WSMO capabilities) can be used as the basis for composition, resulting in a less accurate but presumably more efficient methodology. Both fully automatic methods (without human intervention) and semi automatic methods (with human intervention) shall be considered. Generally, the techniques build upon, and depend crucially on, functionalities provided by other WSMX components, such as discovery, mediation, and reasoning. In particular, composition will typically assume that a set of potential component services is already given (discovered).
A preliminary list of desired functionalities is this:
1. Given a (potentially large) set of web service capability descriptions, and a goal capability, construct a composed service that accomplishes the goal, when ignoring choreographies and assuming that the non statically checkable conditions (WSMO capability assumptions) are met. The composed service may take the form of a sequence of services, or of a partially ordered set of services.
2. Deal with optimization criteria in the above, such as accumulated cost, reputability, or some other notion. Develop methods finding provably optimal solutions, as well as greedy methods that pose a small set of locally optimal solutions as a choice to the user.
3. Given a capability level solution, i.e. a partially ordered (small) set of web services, compose the solution service at the process level.
Suggestions for further functionalities are most welcome!
So far a piece of work was completed in collaboration with IRST Trento (Marco Pistore and Piergiorgio Bertoli), in the context of the KnowledgeWeb project. Discovery, a Reputability mechanism, functional level composition, and process level composition were integrated in a prototype tool. DERI's part here was the functional level composition (FLC), done by Joerg Hoffmann via compilation to a variant of the planning language PDDL. While compilation of FLC to PDDL has been explored by various researchers already, the new aspect of our work here is that FLC was compiled into PDDL with uncertainty. This enabled us to deal with non-trivial (but restricted) ontologies, and with partial matches -- intuitively, the uncertainty was used to express ambiguities as to what concepts an instance belongs to (example: a trip request may request either a flight trip or a train trip, and which one is not known a priori). Future work is to avoid the detour to PDDL, and implement the solution techniques directly in the WSMO context. This will enable us to deal with much more complex background ontologies.
Joerg Hoffmann: functional level composition; for process level composition, a collaboration with IRST Trento (Marco Pistore and Piergiorgio Bertoli) has been established.
James Scicluna: Application and extension of AI Planning techniques for Web Service Composition in the SUPER Context Work Package 3, deliverables d3.2, d3.3
Adina Sirbu: functional level composition in the context of SemBiz Work Package 2