Automating Extraction of Problem Diagrams from Natural Language Requirement Documents

Embedded systems are known for their high complexity and the time cost of manually analyzing and modeling their requirement documents is significantly high. To shorten the time for requirement modeling and reduce the workload of requirements engineers, this paper proposes an automated approach to extract the problem diagram from natural language documents of embedded systems. Specifically, we design neural network models to extract modeling elements from requirements documents and then assembled them into problem diagrams. We conduct experiments on four new datasets collected by this work, using three widely used metrics for evaluation. The experimental results indicate that (1) the approach can extract more correct entity elements, improving 12.99% relative performance compared to the baseline model. (2) The approach is effective to extract the relation elements and the F1 score reached 92.86%. (3) The approach successfully extracts the problem diagram on a real embedded system. Therefore, the approach proposed in this paper can assist in extracting the modeling elements and generating the problem diagram to improve the efficiency of embedding system requirements modeling.