An object detection-based framework for automatically recognizing iStar hand drafts

摘要

Although various requirements modeling tools have been developed in the past decades to facilitate modeling practice, hand drafting is still preferred to a certain extent by some modelers. Thus, digitizing hand drafts has been essential in promoting modeling practice. In this paper, we focus on a requirement modeling language, iStar, which has been investigated for over two decades and has been widely accepted as a valuable method to model and analyze requirements, especially the social aspects. Specifically, we propose a comprehensive framework using customized object detection methods to systematically recognize and digitize iStar hand drafts, promoting the practical adoption of iStar. To evaluate the performance of our approach, we have systematically established a dataset that includes 630 hand drafts. To the best of our knowledge, this is the first iStar hand draft dataset and will contribute to further studies in this field. We have designed and conducted a series of experiments to evaluate our approach based on the data, showing that our approach can effectively digitize iStar hand drafts. The entity recognition rate reached 99.6%. The draft recognition rate reached 91.4%. In addition to the experiments, we have pragmatically applied our approach to actual modeling practice, illustrating the practical applicability of our approach.