An Automated Approach for Identification of Non-Functional Requirements using Word2Vec Model

Abstract

Non-Functional Requirements (NFR) are embedded in functional requirements in requirements specification docu-ment. Identification of NFR from the requirement document is a challenging task. Ignorance of NFR identification in early stages of development increase cost and ultimately cause the failure of the system. The aim of this approach is to help the analyst and designers in architect and design of the system by identifying NFR from the requirements document. Several supervised learning-based solutions were reported in the literature. However, for accu-rate identification of NFR, a significant number of pre-categorized requirements are needed to train supervised text classifiers and system analysts perform the categorization process manually. This study proposed an automated semantic similarity based approach which does not needs pre-categorized requirements for identification of NFR from requirements documents. The approach uses an application of Word2Vec model and popular keywords for identification of NFR. Performance of approach is measured in term of precision-recall and F-measure by applying the approach to PROMISE-NFR dataset. The empirical evidence shows that the automated semi-supervised approach reduces manual human effort in the identification of NFR.