Extração automática de modelos conceituais a partir de requisitos para sistemas autoadaptativos

Abstract

The generation of conceptual models on the Requirements Engineering (ER) stage is a practice that helps on the understanding of the problem and also the initial solution design, providing mechanisms for a visual representation of real-world elements and their relationships. When the software being developed is a Self-Adaptive System (SA), we have situations of uncertainty that must be addressed in the ER stage, making this stage more complex and evidencing the lack of more specific approaches. The generation of these models usually occurs manually, where a team member models the concepts based on preliminary requirements specifications, this is a difficult and complex activity that consumes time and resources, and its results are highly dependent on the modeler’s experience. Based on this, a tool that automates this process would reduce the time spent in the modeling activities, minimize errors, and reduce costs. In this work, we developed a prototype tool based on the Eclipse platform, with the objective of extracting a conceptual model from the textual specifications of requirements for SAs expressed in the RELAX language. We formulate a series of extraction rules that combine Natural Language Processing (NLP) techniques, the RELAX language semantics and the RelaxML UML profile (for RELAX conceptual requirements modeling) to process the requirements and produce a conceptual model as output. We evaluate our outputs with a method based on the area of information retrieval developed specifically for this type of tools, and our results are encouraging, we presented a model elements retrieval (recall) rate at 100% and extras elements rate (over-specification) at 0%. However, the extracted data is not very accurate (accuracy), nearly 65%. Although it is still a prototype, we consider that the objective of the work was achieved, but the results indicate opportunities for improvement.