Finding Component State Transition Model Elements Using Neural Networks: An Empirical Study

摘要

Use cases are popular for writing specifications of a system. However, despite their semi-structured nature, it is often time consuming and error-prone to generate component state transition diagrams from use case documents as it is done manually. While attempts to automate model generation from requirements have increased with the advent of deep neural networks (DNNs), there are limited studies in which a neural network architecture successfully extracts information used to construct a component state transition diagram from use cases. In this paper, we investigate the effectiveness of four different neural network architectures using glove and dependency embeddings to find model elements of component state transition diagrams from use case descriptions. Our results from the study show that we may achieve performance equivalent to humans with F1-scores greater than 0.80 for each model element on test data.