Automatic Detection Method for Software Requirements Text with Language Processing Model
Li Zhang, Liubo Ouyang, Jiahao Qin · 2023
Requirements analysis plays a crucial role in the process of software development. However, due to many factors such as the dichotomy of natural language, requirement texts often present diverse defects, which may lead to anomalies and functional deviations in software systems. In order to solve the above challenges, our study proposes a deep learning-based defect detection method for requirement text, which utilizes deep learning text classification techniques, fully integrates the characteristics of requirement text defects, and designs two defect detection classification models, TextCNN and BERT, for comparative training and performance evaluation based on the construction of a defective text dataset, and constructs a deep learning neural network based on BERT suitable for requirement text defect detection. Experimental results show that the method achieves more than 90% accuracy in defect detection, effectively makes up for the shortcomings of traditional manual review, and significantly improves the efficiency of software development and the quality of requirement text.