AI-Driven User Story Generation

User stories are pivotal in the realm of agile software development, primarily due to their organized structure and the simplicity with which they can be executed. Nevertheless, development teams encounter the arduous task of dealing with the assortment of information expected from diverse sources to manually write user stories. This undertaking not only consumes a significant amount of time but is also susceptible to inaccuracies. We automatically generated user stories relying on two distinct approaches: N-gram representation with linguistic heuristics and the GPT-3 model. We evaluated our work by using diverse user stories corpora and calculating the metrics BLEU, ROUGE-N, and BERTScore for each set of user stories generated by both approaches. For the N-gram approach we achieved an average: ROUGE-N=0.39, BLEU=0.26, BERTScore=0.73; and for GPT-3 model: ROUGE-N=0.46, BLEU=0.27 and BERTScore=0.69. We concluded that, although GPT models excel in producing more comprehensive user stories, N-gram models exhibit a higher level of semantic sensitivity. Considering its simplicity and minimal processing requirements, we recommend employing the N-gram technique for user story generation.