Identification and Classification of Architecturally Significant Functional Requirements

Architecturally Significant Functional Requirements (ASFRs) are those functional requirements that have a significant impact on the architecture of the software system. ASFRs contain comprehensive information to aid architectural decisions; however, their architectural impact is often not explicitly stated in software requirement specification documents. ASFRs are therefore hard to detect, and if missed, can result in expensive refactoring efforts in later stages of software development. Identification and classification of ASFRs using traditional machine learning algorithms have been reported in the past. In this paper, we present our experiments with a deep learning-based model for performing the same task. Our approach is based on a Bidirectional Long Short-Term Memory Network (Bi-LSTM) to capture the context information for each word in the software requirements text, followed by an Attention model to aggregate useful information from these words in order to get the final classification. For ASFR identification, we obtained an f-score of 0.86 and for ASFR classification, we obtained an average f-score of 0. 83.