.png){kind=icon}
.png){kind=icon}
Return
A comparative study on vectorization methods for non-functional requirements classification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.infsof.2022.106991.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Context: Identifying non-functional requirements (NFRs) and their categories at the early phase is crucial for analysts to design software systems and recognize constraints. Automatic non-functional requirements classification methods have been studied for reducing the costs of that labor-intensive task. Our previous study focused on the differences among vectorization methods that converted requirements written in natural language into numerical vectors for classification. It had some limitations regarding the number of datasets used, the types of vectorization methods supporting pre-trained data, and the performance evaluation procedure. Objective: To examine whether different vectorization methods lead to differences in the classification performance of NFRs and their categories with extended settings. Methods: Comparative experiments were conducted with five open data. Nine vectorization methods, including ones with pre-trained data and four supervised classification methods, were supplied. Performance was evaluated with AUC and Scott-Knott ESD test. Results: Some advanced methods could achieve better performance than traditional ones when combined with some classifiers. The use of pre-trained data was useful for some categories. Conclusion: It is beneficial to consider using some combinations of vectorization methods and classifiers for classifying non-functional requirements categories.
Keywords:
Requirements classification
Vectorization methods
Comparative study
Journal
IF:
4.3
Papers:
3.7K
Citations:
7.7K
