Adaptive Web Services Brokering

Abstract

Web services provide a means to architect and operate large-scale distributed information systems. However, syntactic and semantic differences among Web services complicate their interoperability and service composition. Brokers can facilitate their interoperability by providing discovery and mediation services, yet existing approaches are impractical for dynamic applications. We address this limitation by formulating three discovery tasks as supervised learning tasks. In particular, we apply textual case-based and decision tree induction approaches to these tasks and investigate the use of multiple representations. We evaluate their performance in a broker that discovers and mediates requests and responses for meteorological and oceanographic data. Our evaluations show that, for our evaluation tasks, classifiers learned by either approach can effectively perform service discovery in a broker.