Defining Bad Smells and Automating Their Detection in Goal-Oriented Requirement Analysis Method iStar
Yoshitake Hirabayashi, Shinji Ohota, Suzuka Fujii, Motoshi Saeki · 2023
Goal-oriented requirements analysis such as iStar are one of the promising approaches to elicit requirements from stakeholders. However, poor goal models lead to missing requirements and eliciting incorrect requirements as well as less comprehensiveness ones. This paper picks up iStar and proposes a technique to automate detecting bad smells of iStar models, i.e., symptoms of poor models. Firstly, to clarify bad smells, we collected iStar models and developed a list of bad smells. We classified the listed bad smells into two categories: 9 structural bad smells and 4 semantic ones. In the structural bad smells, we focused on the number of graph nodes and the depth of element relationships in iStar models, while in the semantic ones, we used the semantic similarity between descriptions of iStar elements written in natural language. Furthermore, we have developed an automated smell detector for 13 bad smells. Through experiments conducted to evaluate the usefulness of this automated detection tool, our detector could detect 79% of the structural bad smells and 50% of the semantic bad smells, including those that our human subjects overlooked. In addition, the experimental results allowed us to add new 3 bad smells to our smell list.