Large Language Models Capabilities for Software Requirements Automation.

Throughout the las decades, technology has had a very big impact in humans’ life, both positively and negatively, depending on the point of view you prefer to see it. Even though technology may have some negative impacts, related for example to certain types of human jobs that now a days may be substituted with technology, we can say that most of the technological advancements have showed to be of great utility for humans. Everything around us is starting to change with an impacting high dynamism, and software development is very important in order to understand this changes. Moreover, with the recent apparition of the so called artificial intelligence, everything is becoming even more dynamic, and now, humans do not only have on their hands the opportunity to develop software, or execute any kind of technological-related task, but we also have the possibility to automate this tasks, allowing ourselves to do our work in a more efficient way. Efficiency is a word which relates the outcome of the work we can do; with the time we need to spend to do it in the right way. This is exactly the point of study of this investigation. Large Language Models, like ChatGPT, are extremely powerful tools that are already helping millions of people to work efficiently, throughout a widely diverse world of topics. For the subject of our study, we want to show how can LLM’s help human users, with different kind of knowledge, to develop software development requirements, specifically, Use Case Diagrams and Use Case Narratives. Both requirements are extremely useful for software developers and also for the different stakeholders, in order to understand the different corners of the software or application under study. Having correct requirements helps on having also correct effort estimations related to the software to be developed, which is of high importance for the project management organization and financial organization. As mentioned also before, to be efficient, we do not only need to save time on our work, but we also need our work to be well done. To this extent, we developed a detailed methodology on how can we take profit and explode ChatGPT’s capabilities in the best possible way, in order to have correct automated requirements, developed by the AI. To understand this, we need to know that we must think about this like a machine. We are going to have an input, which will be the carefully written prompt introduced to the model, and the we will have a process and an output, related to the LLM. These point are, in a simplistic way, the focal points that we explain throughout the methodology section of the study. Afterwards, since the objective of the work is to analyze the effectiveness of ChatGPT on aiding humans with the automation of the diagrams and narratives, all the results need to be analyzed. For this, a selection of exercises related to both diagrams and narratives are selected and are the object of analysis. The original solution of the exercise can be compared to the generated solution of the AI, and the different components of the requirements can be further analyzed in detail. All these analyses can be found in the results section. Overall, this study is able to demonstrate that ChatGPT, if used throughout the described methodology, is able to showcase very good outcomes, resulting in a useful tool both for the extent of saving time and of having good software development requirements.