Exception detection for web service composition using improved bayesian network

New application systems generated by composition of web services dynamically have become a development trend in network environment. However, since a variety of external services are invoked with different quality of service during processing, the problem of how to keep the execution stable must be addressed in order to improve the reliability and availability of the combination of services. In this paper, an approach of exception detection is presented for web service composition by improved Bayesian network. Firstly, the topological structure of Bayesian network is established, where the causal relationships between underlying web services are mapped to Bayesian network in service combination process. The conventional Bayesian network algorithm is improved to satisfy the demand of better conversion from nodes in web service composition to nodes in Bayesian network. Secondly, the parameter setting in Bayesian network is explored for determining the prior probability of service nodes and the conditional probability of service output node. Thirdly, the algorithm of the improved exception detection model is introduced, with the aim of locating the web services that cause exceptions in the execution of composite services. Lastly, a case study is given and the results analysis is also conducted. The results show that the presented approach not only considers the uncertainty during exception detection, but also can identify the exception web services in the process.