Identification of user stories in software issues records applying pre-trained natural language processing models

Abstract

In the last decades, agile development methods have been increasingly adopted by the software industry. User stories are one of the primary development artifacts for agile project teams. Issue Management Systems are widely used by software development teams to generate user stories, and organize them in meaningful fragments: epics, themes, and sprints. In addition, these tools enable generating any kind of issues, like bugs, change requests, tasks, etc. The responsibility for correctly categorizing an issue is in the hands of the team members, so it is a task prone to errors and frequently omitted due to lack of time or bad practices. Thus, a current problem is that many issues in projects remain uncategorized or mislabeled. Several studies have shown that it is common to find the uncategorized user stories of a software project in large volumes of issues records maintained by Issue Management Systems. In this work, we present two Neural Network models for text classification that were implemented for the identification of user stories in issue records.