Discovery of Optimized Web Service Configurations Using a Hybrid Semantic and Statistical Approach

We present a semantic optimized service discovery (SemOSD) approach capable of handling Web service search requests on a fine-grained level of detail where we augment semantic service descriptions with statistically built predictor functions. Our approach combines ontologies and mathematical functions built using statistical regression over previous Web service interactions. In the search requests we allow for arbitrary, independent and dependent constraints and user preferences expressed using objective functions. Our approach maps to standard operational research global optimization problem where algorithms of simulated annealing and differential evolution are used. It is capable of finding the optimal combination of service input and output parameters (a configuration) to a user request with rich preferences. Our approach is applied to an international package shipment scenario where real (Web) services are used and mined to create price prediction models. We show that the chosen regression method provides price prediction models of high accuracy and our approach supports expressive and complex search requests.