Many of the previous methods for composing Web processes utilize either classical planning techniques such as hierarchical task networks (HTNs), or decision-theoretic planners such as Markov decision processes (MDPs). While offering a way to automatically compose a desired Web process, these techniques do not scale to large processes. In addition, classical planners assume away the uncertainties involved in service invocations such as service failure. In this paper, we present a hierarchical approach for composing Web processes that may be nested - some of the components of the process may be Web processes themselves. We model the composition problem using a semi-Markov decision process (SMDP) that generalizes MDPs by allowing actions to be temporally extended. We use these actions to represent the invocation of lower level Web processes whose execution times are uncertain and different from simple service invocations.
展开▼
