Experiment on automatic functional requirements analysis with the EFRF's semantic cases

Abstract

Software requirements analysis is crucial for any software project and it is the basis of requirements reuse within Software Product Line engineering. Software requirements specifications are usually expressed in natural language, which are informal, imprecise and ambiguous, thus analyzing them automatically is a challenging task. Although methods towards automatic analysis of software requirements have been studied before, many of them have limitations and effective researches in this area are still lacking. Therefore, in this paper a new approach was proposed to automatically extract structured information of functional requirements from Software Requirements Specifications in natural language. The methods of machine learning, natural language processing and semantic analysis were employed and combined in this approach. With a 10-fold cross validation setting, the method was evaluated on a manually annotated corpus. The experiments show that this approach achieves a good performance. Moreover, it was found that the model trained on the requirements dataset of the E-commerce systems, can be used to extract semantic information from the requirements of auto-maker systems. The cross-domain extraction results show that the method of this paper is domain independent and robust to some extent.