A Method for Selecting Training Data Using Doc2Vec for Automatic Test Cases Generation
Yuto Fujita, Kiyoshi Ueda · 2024
In the development of large-scale communication software, due to the increase in development cost and shortage of labor, a method to automatically generate test cases of system testing and acceptance testing from requirement specification documents using machine learning has been studied. In this study, we improve the accuracy of automatic test cases generation by selecting requirement specification documents for machine learning training data. We studied a method to select structured requirement specification documents with high similarity to the requirement specification document of test data and use them as training data. We propose a method that vectorize each requirement specification document using Doc2Vec and calculate the similarity between each requirement specification document vector. We evaluate the effectiveness of the proposed method by measuring the accuracy using the proposed method on requirement specification documents of large-scale communication software. Experiments were also conducted on different systems requirement specification documents to confirm the effectiveness of the method on a wide variety of requirement specification documents.