Log-based Mining Techniques Applied to Web Service Composition Reengineering

The original publication is available at www.springerlink.comInternational audienceThere is no doubt that SOA and BPM will continue to evolve dependently for the next ten years. Preparing common research infrastructures will require most important efforts of web service community researchers. Web services composition is becoming more and more chaotic, involving numerous interacting ad-hoc services through huge business processes. Analysing and Reengineering of complex Web Service Compositions will enable them to be well understood, controlled, and redesigned. Our contribution to this problem is a patterns mining algorithm which is based on a statistical technique to discover patterns from execution log. Our approach is characterised by a “local” patterns discovery that allows to cover partial results through a dynamic programming algorithm. Those local discovered patterns are then composed iteratively until discovering the Composite Web Service. The analysis of the disparities between the discovered model and the initial ad-hoc CS model (delta-analysis) enables initial design gaps to be detected and thus the Web service composition to be re-engineered