Accuracy Improvement by Training Data Selection in Automatic Test Cases Generation Method

Abstract

The development and maintenance costs of the high-quality communication software tend to remain high because the telephone services must be reliable and secure as they are valuable social infrastructure. Previous studies formulated the description style of the requirements specification in a form with which a machine could deal and had customers use this style to describe the requirements specification. However, no method has been developed to generate test cases from natural language documents. The method for automatically generating the test cases of system testing and acceptance testing from the requirement specification is studied. We propose training data selection quality improvement technique in the cosine similarity with the test data. We confirmed the effectiveness of the methods. We also proposed second method adding the application judgment technique by the standard deviation value. We confirmed usefulness of the proposed methods that obtain the maximum value of accuracy with less training data.