Automated Identification of Component State Transition Model Elements from Requirements

Most system requirements are currently written in common, i.e., unstructured, natural language, which existing requirements analysis tools are poorly equipped to handle. Extracting mentions of model elements from common natural language requirements is a first step toward the automation of model-driven requirements analysis. We propose an approach in which we identify mentions of elements of a component state transition (CST) model in natural language requirements by creating classifiers using a recurrent neural network with long short-term memory. To evaluate our approach, we performed a study on a pacemaker system requirements document, and the results show promising directions for future research.