Text mining support for software requirements: Traceability assurance

Requirements assurance aims to increase confidence in the quality of requirements through independent audit and review. One important and effort intensive activity is assurance of the traceability matrix (TM). In this, determining the correctness and completeness of the many-to-many relationships between functional and non-functional requirements (NFRs) is a particularly tedious and error prone activity for assurance personnel to peform manually. We introduce a practical to use method that applies well-established text-mining and statistical methods to reduce this effort and increase TM assurance. The method is novel in that it utilizes both requirements similarity (likelihood that requirements trace to each other) and dissimilarity (or anti-trace, likelihood that requirements do not trace to each other) to generate investigation sets that significantly reduce the complexity of the traceability assurance task and help personnel focus on likely problem areas. The method automatically adjusts to the quality of the requirements specification and TM. Requirements assurance experiences from the SQA group at NASA's Jet Propulsion Laboratory provide motivation for the need and practicality of the method. Results of using the method are verifiably promising based on an extensive evaluation of the NFR data set from the publicly accessible PROMISE repository.