A Clustering Technique for Early Detection of Dominant and Recessive Cross-Cutting Concerns

This paper describes an approach for automating the detection of early aspects. Based on hierarchical clustering and an underlying probabilistic algorithm, the technique generates initial requirements clusters representing relatively homogenous feature sets, use cases and potential cross-cutting concerns. A second clustering phase is then applied in which dominant terms are identified and removed from each of the initial clusters, allowing new clusters to form around less dominant terms. This second phase enables previously inter-tangled aspects to be detected. Three metrics are introduced to differentiate potential cross-cutting concerns from other types of clusters. The approach is illustrated through an example based on the Public Health Watcher case study.