Temporal relation identification in functional requirements

In this study, we propose a method for applying a temporal relation identification model to functional requirements. We discuss the limited availability of data in the requirements engineering domain compared to other fields when used for supervised learning, and therefore employ a corpus from the news domain for training. The experimental results demonstrate that the types of temporal relations present in functional requirements are limited, indicating that focusing on learning with a narrowed set of labels is effective. Additionally, We incorporate Dependency Path (DP) into the temporal relation identification model and report, through comparative experiments, that leveraging DP is effective, but minor modifications to DP do not lead to significant improvements in accuracy. By demonstrating specific application methods of temporal relation identification in requirements engineering, we anticipate contributing to the analysis of functional requirements in software development.