Adaptive Process Execution in a Service Cloud: Service Selection and Scheduling Based on Machine Learning

Given a process specification, it is a complex task to dynamically select constituent services and compose them in an execution plan to satisfy users´ non-functional preferences. Process scheduling approaches assume users can clearly specify their non-functional preferences and there are formulas (e.g., utility functions) to compute process level QoS from the QoS of constituent services and their connections. However, these assumptions are not always true. Users´ preferences can be subjective, implicit, vague, mixed and different for various types of processes. Besides, not all the preferences for example easy-to-use can be computed using formulas. We proposed a machine learning based approach to evolutionarily learn user preferences according to their ratings on historical execution plans, recommend existing or generate new execution plans for business processes that adapt to user preferences.