Discovering Relationships Among Software Artifacts

摘要

Software systems have become ubiquitous in today's world. Most software will evolve after initial deployment. Software changes that are a part of that evolution often are documented in a requirements change document. One of the challenges when changing software is understanding the portions of the existing requirements and the existing code that could be affected by the change in order to avoid or minimize unexpected side effects from the changes. Researchers have addressed the problem of minimizing the effect of changes by using different methods, including text mining and clustering. Some approaches to determine change impact are based on information retrieval (IR) techniques using both term frequency-inverse document frequency (TF—IDF) and latent semantic indexing (LSI) methods. Other approaches are based on visualization techniques using degree and betweenness centrality measures. In this research, we approach the problem by applying IR techniques along with data mining. We apply TF-IDF and LSI to investigate which changes have a high potential of modifying existing requirements. We also analyze similarities between changes that do not map to existing requirements. In both cases, our threshold for identifying similarity is 80%. We designed our approach to identify, for a given change, one or more requirements that have a high potential of being associated with the change as well as identifying intra-document requirements or changes that have a high potential for consolidation. We were able to identify requirements that had a similarity of at least 80% to a change request using TF-IDF and LSI. We were also able to isolate changes that did not show a high similarity to any requirement, thus indicating that the change request was likely a request for a new requirement. The results are encouraging for assessing the impact of software change requests on requirements of an existing system.