Automatic user story generation: a comprehensive systematic literature review

User stories are the lifeblood of agile software development due to their semi-structured format and ease of implementation. However, the variety of data sources and textual formats used to document software requirements bring a challenge for software development teams. They often need to read and comprehend the client’s needs from different sources and convert them into user stories manually. This process demands time, and it is also prone to errors. As an alternative to remedy this issue, there are studies concerning the automatic generation of user stories. We conducted a systematic literature review (SLR) to identify and analyze existing approaches for automatically generating user stories. We investigated which type of corpora were used for training and testing, which Natural Language Processing (NLP) or Machine Learning (ML) techniques are employed to reach this goal, and how researchers are evaluating the quality of the user stories generated. Our SLR followed established guidelines and investigated state-of-the-art research from prominent academic publishers such as ACM, IEEE Xplore, and ScienceDirect. Studies published until April 2024 were included, with a focus on those addressing the research questions proposed. Our findings indicate a critical shortage of publicly available corpora hindering advancements in this field, especially in the current era of ML. The team also found there is a broad variety of techniques being employed on this topic. Finally, the studies need to pay more attention to guidelines for evaluating user stories quality. The automatic user story generation remains in its early stages. We highlight some opportunities for contribution and discuss the direction of future works.