CRP: context-based reputation propagation in services composition

For a number of services with similar functionality reputation has been regarded as one of the most important methods to identify good ones from bad ones. However, a composite service, which is composed of multiple component services, obtains only one score (or feedback) after every invocation. In order to compute the reputation of each component service, it is necessary for the composite service to distribute this score to its component services. How to achieve a fair distribution is a challenging issue, as each component service may perform differently in contributing to the success or failure of the composite service. Although several efforts have been made for this problem, they do not consider the context of composition, which makes the distribution unfair. Therefore, in this paper, we propose a fair score distribution framework which combines the context of component services and their runtime performance. We distinguish two aspects contexts of a component service: structure-related importance and community-related replaceability, and adopt graph theory and dominating relationship technique to compute them, respectively. Experimental results show that our approach can achieve a more reasonable and fair score distribution than other existing methods.