Effective automatic dynamic semantic Web service composition FIIT - 10890 - 4520 Dissertation

Web services are nowadays a popular technology used in practice when diverse software system integration is in demand. The ability to make some functionality available through the Web has inspiring consequences. The Web services provide features which are very interesting also from a research point of view. They are studied from several aspects to propose solutions allowing their more sophisticated exploitation. One of the intensively researched areas is the study how Web services can be used to dynamically create a functionality, based on the actual requirements. The basic idea is that multiple Web services can be combined together, to form a composite service, supplying more complex needs. The particular composition is realized automatically, on the fly, based on the actual goal. To facilitate this kind of Web service utilization, additional metadata depicting the functionality of single services is required. These metadata are provided in a form of semantic annotations. This thesis deals with automatic dynamic semantic Web service composition. It discusses the problem of service composition in a wider context and deals with selected sub-problems concerning this topic. The sub-problems include the proper description of the behavior of Web services, management of the changes in the service environment, and handling multiple composition requests. In the context of service behavior description, we precisely define the pre-/post-conditions of Web services and the rules followed when these are considered during the composition. We show that it is feasible to perform pre-/post-condition aware composition. The question if service composition is realizable is studied also from the point of view of large-scale repositories including a huge number of Web services. We show that composition can be performed in an acceptable time even if repositories consisting of several ten-thousands of services are considered. This is true even if optimization according to the non-functional properties of the services is required. We also show that the changes in the service environment, such as adding/removing a service, or a change of the values of non-functional properties, can be effectively managed. Finally, we show that effective service composition is possible also in real scenarios where multiple composition requests arrive independently from several users.