Semantic Web Service Composition via Logic-based Program Synthesis

The ability to efficient selection and integration of inter-organizational heterogeneous Web services at runtime becomes an important requirement to the Web service provision. In an Web service application, if no single existing Web service can satisfy the functionality required by the user, there should be a program or an agent to automated combine existing services together in order to fulfill the request. The aim of this thesis is to consider the Web service composition problem from the viewpoint of logic-based program synthesis, and to propose an agentbased framework for supporting the composition process in scalable and flexible manner. The approach described in this thesis uses Linear Logic-based theorem proving to assist and automate automatic composition of Semantic Web services. The approach uses a Semantic Web service language (DAMLS) for external presentation of Web services, while, internally, the services are presented by extralogical axioms and proofs in Linear Logic. Linear Logic, as a resource conscious logic, enables us to capture the concurrent features of Web services formally (including parameters, states and non-functional attributes). The approach uses a process calculus to present the process model of the composite service. The process calculus is attached to the Linear Logic inference rules in the style of type theory. Thus the process model for a composite service can be generated directly from the complete proof. We introduce a set of subtyping rules that defines a valid dataflow for composite services. The subtyping rules that are used for semantic reasoning are presented with Linear Logic inference figures. The composition system has been implemented based on a multi-agent architecture, AGORA. The agent-based design enables the different components for Web service composition system, such as the theorem prover, semantic reasoner and translator to integrated to each other in a loosely coupled manner. We conclude with discussing how this approach has been directed to meet the main challenges in Web service composition. First, it is autonomous so that the users do not required to analysize the huge amount of available services manually. Second, it has good performance so that the composition is performed in a dynamic environment. Third, it solves the heterogenous problem because the Semantic Web information is required for matching and composing Web services. We argue that LL theorem proving, combined with semantic reasoning offers a flexible approach to the success to the composition of Web services. LL, as a logic for specifying concurrent programming, provides higher expressive powers to model Web services than classical logic. Further, the agent-based design enables the different components for Web service composition system to integrated to each other in a loosely coupled manner. The contributions of this thesis is summarized as follows. First, an generic