| Abstract [eng] |
In software development companies, a requirements list is often prepared before creating a commercial proposal and signing a contract, but this process is complex, time-consuming, and costly, one of such methods is EventStorming workshop. The analysis of existing solutions showed that there are no sufficiently developed solutions that would enable automated requirements extraction specifically from EventStorming artifacts. Therefore, this thesis aims to facilitate requirements elicitation by automating the transformation of EventStorming artifacts into structured requirements. The thesis defines the requirements for the proposed tool, introduces a methodology for automated requirements extraction from EventStorming artifacts, and presents the implemented tool. The tool consists of five main components: artifact recognition, text recognition, artifact clustering, requirements generation, and requirements export. Sticky note detection was implemented using the SSD MobileNet V1 FPN architecture, text recognition was performed using Google Vision API OCR, and requirements generation was carried out by instruction-tuned agent using large language model. The developed solution was evaluated in three experiments. In the sticky note detection experiment, the model achieved 0.9626 precision, 0.8869 recall, and an F1-score of 0.9232. In the artifact type recognition experiment, the overall accuracy reached 94.03%. The practical usefulness of the tool, the generated user stories, and the overall process was evaluated through a respondent survey, where the mean scores of the three evaluated statements ranged from 4.44 to 4.56 out of 5, and 89.3% of all ratings were either 4 or 5. The results show that the developed tool can facilitate requirements extraction from EventStorming artifacts and is practically useful in real-world projects. |
