<![CDATA[Requirement engineering has a critical role in software development. Various techniques have been invented to improve requirements quality. Recently, natural language processing (NLP) has drawn attention to the software engineering community, due to its capacity for automated requirements elicitation which accelerates software development. However, NLP through machine learning approaches has been discovered to be ineffective, caused by the variability in natural language, unbalanced dataset, heavy preprocessing tasks, and issues in hyperparameter tuning. A deep learning approach on the other hand has been found to handle enormous dataset without sophisticated preprocessing task and to improve classification quality for unstructured data such as images and text. This paper investigates how automated software requirements classification can be improved and how well deep learning approaches work. We aim to contribute an approach to minimize preprocessing requirements to improve classification accuracy from OpenScience tera-PROMISE repository. We also compare our approach with several techniques that have been previously tested. We find that our technique may improve the performance of an existing classification method. Finally, we present significant differences in the performance of approaches, such as for the sub-classification of nonfunctional requirements.]]>
Copyrights © 2023
