Predicting change impact in Web service ecosystems
The paper aims to address the issue of Web service providers facing a major issue of estimating the potential effects of changing a Web service to other services, especially in large ecosystems of Web services which have become more common nowadays. Web service providers make constant changes to their Web services to meet the ever-changing business requirements. Design/methodology/approach - The paper proposes an approach to predict change impact by mining a version history of a Web service ecosystem. The proposed approach extracts patterns of Web services that have been changed together from the version history by using association rule data mining techniques. The approach then uses this knowledge of co-changed patterns for predicting the impact of future changes based on the assumption that Web services which have been changed together frequently in the past will likely be changed together in future. Findings - An empirical validation based on the Amazon's ecosystem of 46 Web services indicates the effectiveness of the proposed approach. After an initial change, the proposed approach can correctly predict up to 25 per cent of further Web services to be changed with the precision of up to 82 per cent. Originality/value - Traditional approaches to predict change impact in Web services tend to rely on having a dependency graph between Web services. However, in practice, building and maintaining inter-service dependencies that capture the precise semantics and behaviours of the Web services are challenging and costly. The proposed approach offers a novel alternative which only requires mining the existing version history of Web services.