LEARNING-BASED WEB SERVICE COMPOSITION IN UNCERTAIN ENVIRONMENTS

Web service composition has two kinds of uncertain factors, including uncertain invocation results and uncertain quality of services. These uncertain factors affect success rate of service composition. The web service composition problem should be considered as an uncertain planning problem. This paper used Partially Observable Markov Decision Process to deal with the uncertain planning problem for service composition. According to the uncertain model, we propose a fast learning method, which is an uncertainty planning method, to compose web services. The method views invocations of web service as uncertain actions, and views service quality as partially observable variables. The method does not need to know complete information, instead uses an estimated value function to approach a real function and to obtain a composite service. Simulation experiments verify the validity of the algorithm, and the results also show that our method improves the success rate of the service composition and reduces computing time.