Efficient Service Variant Analysis with Markov Updates in Monte Carlo Tree Search (Short Paper)

Static analysis techniques can be used to analyse and simplify interfaces of enterprise systems, such as those from SAP, Oracle and FedEx, which becoming more prominent on the internet and vying for new systems integration and extension opportunities. Web services of enterprise systems are notoriously complex, having hundreds of parameters per operation, multiple levels of nesting, leading to ambiguities about valid invocations of operations. To derive valid invocations, which in turn assists service users with invoking services correctly, this paper focuses on a challenging aspect of static interface analysis, namely, the identification of service variants in operations, in which the parameters are subtypes of business entities involved in a service. To efficiently search for which combinations of parameters are for a valid invocation, we have proposed a Monte Carlo method, based on likelihood-free Bayesian sampling, to identify higher probability parameters spaces, from which to test prospective invocations. A significant performance boost was found by extending Monte Carlo sampling with Markov look-up, with validation using a simulated FedEx service interface, whose structural complexity exceeds many web services of enterprise systems available on the internet.