Análise comparativa de ferramentas para detecção de anomalias de requisitos de software

Abstract

A software requirement indicates a capability or a characteristic that a software must have in order to be of value to its stakeholders. It is essential to ensure that the description of the requirements is clear and unambiguous, in order to allow their correct understanding and facilitate their evolution.However, as most software requirements are described in natural language, they may contain subjectivities and inconsistencies in their description, which is called “Software Requirements Anomaly”. Several works have proposed tools with the objective of contributing to the detection of requirements anomalies. However, it can be noted that few of these works have evaluated the effectiveness (coverage and accuracy) of the proposed tools, comparing them with each other and trying to identify their strengths and weaknesses. Therefore, this work, applied to perform an experimental study to cover this gap in the literature. In all, 3 (three) anomaly detection tools were carefully selected from the literature (RETA, Tactile Check and Tiger Pro) and analyzed based on requirements documents from different domains, containing more than 85 anomalies. As a result, the tool with the highest coverage is RETA, having correctly identified 85.19% of the existing anomalies. Nevertheless, considering the average between coverage and accuracy, the Tactile Check tool was the one that presented the best result. Despite this, all the tools comparatively analyzed in this work had an unsatisfactory level of coverage and accuracy, falling below 66% and 57% respectively. This is largely due to the words that make up the dictionary not being well calibrated with the anomalies that the tool is willing to detect and the detection technique not considering the context of the description of the requirement of which the anomalous term is inserted.