A clustering-based approach for discovering flaws in requirements specifications

In this paper, we present the application of a clustering algorithm to exploit lexical and syntactic relationships occurring between natural language requirements. Our experiments conducted on a real-world data set highlight a correlation between clustering outliers, i.e., requirements that are marked as "noisy" by the clustering algorithm, and requirements presenting "flaws". Those flaws may refer to an incomplete explanation of the behavioral aspects, which the requirement is supposed to provide. Furthermore, flaws may also be caused by the usage of inconsistent terminology in the requirement specification. We evaluate the ability of our proposed algorithm to effectively discover such kind of flawed requirements. Evaluation is performed by measuring the accuracy of the algorithm in detecting a set of flaws in our testing data set, which have been previously manually-identified by a human assessor.