Edinburgh Napier University

  Automatic classification of software requirements is an active research area; it can alleviate the tedious task of manual labeling and improves transparency in the requirements engineering process. Several attempts have been made towards the identification and classification by type of functional requirements (FRs) as well as non-functional requirements (NFRs). Previous work in this area suffers from misclassification. This study investigates issues with NFRs in particular the limitations of existing methods in the classification of NFRs. The goal of this work is to minimize misclassification and help stakeholders consider NFRs in early phases of development through automatically classifying requirements. In this study, we have proposed an improved requirement detection and classification technique. The following summarizes the proposed approach:

(a) A newly created labelled corpus,

(b) Textual semantics to augment user requirements by word2vec for automatically extracting features, and

(c) A convolution neural network-based multi-label requirement classifier that classifies NFRs into five classes: reliability, efficiency, portability, usability, and maintainability.