Detecting Use Case Scenarios in Requirements Artifacts: A Deep Learning Approach.

Use case scenarios (UCS) written in natural languages like English are popular tools for requirements elicitation. They also play an important role in the model driven design process by being used as an initial input for many automated behavior modeling such as generating sequence diagrams and class diagrams. However, there is no unified approach used by the engineers to represent use cases and make it hard to identify use case statements from requirements artifacts without doing it manually by human experts, which is time consuming and tedious. In this paper, we propose a novel approach for automatically identifying use case scenarios within requirements documents written in English. We employ a machine learning based approach which is applicable to a wide range of specifications in various domains. We train and evaluate different state-of-the-art machine learning models including transformers over an independently labeled dataset to compare the performance of different models and select the best classifier based on the result. Our experimental result shows that ULMFiT outperforms other models with an accuracy of 95%. Moreover, we disclose our dataset as well as the source code to foster the discourse on the automatic use case detection within the research community.