Effective service composition in large scale service market: An empirical evidence enhanced approach

The web has undergone a tremendous shift from information repository to the provisioning capacity of services. As an effective means of constructing coarse-grained solutions by dynamically aggregating a set of services to satisfy complex requirements, traditional service composition suffers from dramatic decrease on the efficiency of determining the optimal solution when large scale services are available in the Internet based service market. Most current approaches look for the optimal composition solution by real-time computation, and the composition efficiency greatly depends on the adopted algorithms. To eliminate such deficiency, this paper proposes a semi-empirical composition approach which incorporates the extraction of empirical evidence from historical experiences to provide guidance to solution space reduction to real-time service selection. Service communities and historical requirements are further organized into clusters based on similarity measurement, and then the probabilistic correspondences between the two types of clusters are identified by statistical analysis. For each new request, its hosting requirement cluster would be identified and corresponding service clusters would be determined by leveraging Bayesian inference. Concrete services would be selected from the reduced solution space to constitute the final composition. Timing strategies for re-clustering and consideration to special cases in clustering ensures continual adaption of the approach to changing environment. Instead of relying solely on pure real-time computation, the approach distinguishes from traditional methods by combining the two perspectives together.