In the development of large-scale communication software, due to increasing development costs and labor short-ages, we have been studying methods to automatically generate test cases for system testing and acceptance testing from requirement specifications using machine learning. In this paper, we establish a method for automatic test case generation by automatically structuring requirement specifications using deep learning, and aim to further improve its accuracy. The accuracy of the proposed method is compared with the CRFs of previous studies for the requirements specifications of session control servers and various IT systems, and its effectiveness is confirmed.