Automatic Test Cases Generation with Selection of Training Data for Various System Specifications
Kiyoshi Ueda, Riku Shikama, Yuki Shimizu · 2024
We have studied methods to automatically generate test cases from requirement specifications using machine learning for high-quality software development. In this study, we improve the accuracy by learning on structured requirement specifications similar to the new requirement specification document for which we want to generate test cases. We propose a method to vectorize requirement specifications using a topic model (LDA) and select structured requirement specification documents with high cosine similarity to the new requirement specification document vector as training data. We applied the proposed method to requirement specifications of various real IT systems and evaluated the accuracy of the input and output information on a test case basis. We confirmed an accuracy of about 85% and confirmed that the proposed method can be used for test case generation. We also confirmed that the accuracy is high when the number of requirement specifications included in the training data is small, and that the quality of the training data improves.