Master Thesis "Generating-Test-Cases-Using-NLP" https://lup.lub.lu.se/student-papers/search/publication/9168588
Software testing today is a vital factor in maintaining the quality and reliability of software products in an always-advancing technical era. Nevertheless, making manual high-quality test documents has been historically laborious and lengthy, which results in significant time and money consumption for the whole software creation process. Recently, the use of transformer models has become an efficient tool for the automation of this procedure. This thesis is based on the use of large language models to create test case documents from feature specifications written in natural language. We assess different ways to improve our model’s effectiveness, such as fine- tuning, prompt engineering, and agentic workflow methods. We carry out the research with the use of a quantized and optimized model for memory efficiency that demonstrates the possibility of generating good test cases even with the re- striction of computational resources. With our approach, we achieved remark- able results in both the BLEU score and the human evaluation score. Our highest BLEU score achieved with our best approach is 32.93, which also corresponded to the highest human evaluation score 3.71. This is not far from the reference value 4.68 of being humanly written and undergoing a review process.
