BibSonomy ::  publication ::

In Web Service Composition WSC problems, the composition process generates a solution, i.e., a composition or a plan of atomic services, whose execution achieves some objectives on the Web. Existing research on Web service composition generally assumed that these objectives are absolute; i.e., the service-composition algorithms must achieve all of them in order to generate successful outcomes; otherwise, the composition process fails altogether. The most straightforward example is the use of OWL-S process models that specifically tell a composition algorithm how to achieve a functionality on the Web. However, in many WSC problems, it is also desirable to achieve users' preferences that are not absolute objectives, but a solution composition generated by a WSC algorithm must satisfy those preferences as much as possible. In this paper, we first describe a way to augment OWL-S process models by qualitative user preferences. We achieve this by mapping a given set of process models and preferences into a planning language for representing Hierarchical Task Networks HTNs. We then present SCUP, our new WSC algorithm that performs a best-first search over the possible HTN-style task decompositions by heuristically scoring those decompositions based on ontological reasoning over the input preferences. Finally, we discuss our theoretical and experimental results on the SCUP algorithm.