Test case generation using an Attention Seq2Seq model for structuring requirement specifications

Abstract

In the development of large-scale communication software, various methods have been investigated to address challenges such as increasing development costs and shortages of skilled personnel by leveraging machine learning to automatically generate system test cases from requirement specifications. To further improve the accuracy of automatically generated test cases, this study proposes a method that employs deep learning to structure requirement specifications and generate test cases. In particular, we introduce an Attention Seq2Seq model. The proposed model achieved significantly higher accuracy than the LSTM model presented in previous studies.