Isolating and relating concerns in requirements using latent semantic analysis

Abstract

Aspect-oriented requirements analysis involves the identification of concerns that behaviorally influence other concerns. Such concerns are described in requirements called emphaspectual requirements: requirements that detail the influence of one concern over another. The current state of the art for aspect-oriented requirements analysis is Theme/Doc, which allows lexical analysis of requirements based on a set of developer-chosen keywords. It provides a graphical depiction of how concerns relate to requirements, and affords identification of potential aspectual requirements. In addition, clusters of requirements and concerns are identified to arrive at a more useful set of concerns than those initially identified.Because of the lexical nature of the Theme/Doc approach, aspectual requirements are missed, or wrongly identified. Additionally, requirements may be wrongly clustered if they contain ambiguous terms.In this work we explored whether the use of a statistical approach for textual analysis, Latent Semantic Analysis (LSA), would improve upon the lexical approach used by Theme/Doc. We found that LSA helps identify useful concern clusters, and helps reduce the number of falsely identified aspectual requirements.