An Attention Based LSTM Model: Automated Requirement Classification from User Story

A significant obstacle in Software Requirement Specification (SRS) is the manual classification of software requirements, which can be a subjective and time-consuming endeavor. It can result in inconsistency and the possibility of classification errors in the requirement collection process. Moreover, as the number of requirements rises, the required manual labor becomes increasingly difficult and impracticable. To surmount these obstacles, therefore, the classification of software requirements must be automated. Hence, in this research, an attention-based LSTM model is proposed, for automatic classification with the goal of addressing the significance of automation in software requirement classification. An experimental study has been conducted on two datasets containing nearly 1,039 Functional and Non-Functional Requirements (FR, NFR). The experimental results demonstrate that the proposed model classifies FR and NFR with an accuracy of 98% on the testing dataset. The outcome underscores the effectiveness of utilizing attention-based LSTM models for requirement classification and alleviates the ambiguities and challenges associated with manual classification, ultimately improving the effectiveness and reliability of software requirement analysis.