Un marco conceptual para la generación de requerimientos de software de calidad

Requirements expressed in natural language are an indispensable artifact in the software development process, as all stakeholders can understand them. However, their ambiguity poses a persistent challenge. To address this issue, organizations such as IEEE and INCOSE publish guidelines for writing requirements, offering rules that assist in this task. On the other hand, agile methodologies provide patterns and structures for expressing stakeholder needs in natural language, attempting to constrain the language to avoid ambiguity. Nevertheless, the knowledge gap among stakeholders regarding the requirements and the correct way to express them further complicates the specification task. In recent years, large language models (LLMs) have emerged to enhance natural language processing tasks. These are Deep learning-based architectures that emulate attention mechanisms like those of humans. This work aims to test the demonstrated power of LLMs in this domain. The objective is to use these models to improve the quality of software requirements written in natural language, assisting analysts in the requirements specification. The proposed framework, its architecture, key components, and their interactions are detailed. Furthermore, a conceptual test of the proposal is developed to assess its usefulness. Finally, the potential and limitations of the framework are discussed, along with future directions for its continued validation and refinement.