Test description generation from software requirements and performance analysis: Application of advanced NLP models

Software testing is of critical importance in the software development life cycle. It is a process that verifies whether software provides the specified functionality, performance, and quality. Software testing enables early detection of errors, providing opportunities to intervene at a stage when their correction is more feasible, thus preventing unnecessary interruptions in the development process. In addition, well tested software reduces extended support and maintenance costs for the customer, and long-term use result in lower maintenance costs and legal risks. In other words, software testing increases the quality of software while reduces costs. However, performing tests alone is not enough for an effective testing process; it requires quality test descriptions that detail the tests. Test descriptions standardize testing processes, record possible test scenarios, and present expected test results. Writing a test description is a time-consuming process that requires expertise and is open to human error. Therefore, it has become a necessary for test descriptions to be automatically generated independently of the person, quickly, and accurately. In this thesis, it is aimed to produce test descriptions from software requirements in a faster and higher quality manner using NLP models, which are among the pre-trained models in the literature that have been fine-tuned with the prepared datasets. As a result of the studies and analyses, it has been observed that it is possible to create a test description from software requirements faster and higher-quality.