Towards Fully Automated Semantic Web Service Composition Based on Estimation of Distribution Algorithm

Web service composition has been a challenging research area, where many researchers have been working on a composition problem that optimizes Quality of service and/or Quality of semantic matchmaking of composite solutions. This NP-hard problem has been successfully handled by many Evolutionary Computation techniques with promising results. Estimation of Distribution has shown its initial promise in solving fully automated service composition, and its success strongly relies on distribution models and sampling techniques. Our recently published work proposed a Node Histogram-Based approach to fully automated service composition. However, many services presented in sampled optimized queues does not contribute to decoded solutions of the queue. Therefore, efforts should be made to focus on learning distributions of component services in solutions. Consequently, we aim to learn more suitable distributions considering services satisfying service dependency in the solutions and use the Edge Histogram Matrix to learn restricted sampled outcomes satisfying the dependency. Besides that, we proposed effective sampling techniques with high efficiency in a straightforward implementation. Our experimental evaluation using benchmark datasets shows our proposed EDA-based approach outperforms two recent approaches regarding both efficiency and effectiveness.