Designing a Pipeline to extract User Stories from Requirements Engineering Meeting Transcripts

Abstract

[Context and Motivation] The quality of Requirements Engineering meetings and the resulting requirements correlate with the quality of a software product. These requirements can be written in the standardized form of a User Story. [Question/Problem] It takes time to manually extract these requirements from the meeting transcripts and write these down as User Stories. The thesis goal is to explore how to automatically extract User Stories from software Requirements Engineering (RE) meeting transcripts. [Principal ideas/results] A Transcript to User Story Pipeline (TUSP) is designed. This pipeline combines already existing Requirement Specification Algorithms (RSAs). The TUSP also uses a novel RSA, specifically designed for this thesis, called the Fit-Gap Searcher. The TUSP consists of 2 configurations: a Machine Learning Configuration and a Lexical Configuration. These two configurations are validated qualitatively and quantitatively on two real-world test cases: a Greenfield test case and a Customization test case. Both configurations predict a couple of good quality User Story fragments, but no whole User Story of good quality is found. The configurations are assessed on the precision, accuracy, recall and F1 metrics, all of which score no higher than 0.1 for both of the configurations. [Contribution] This thesis presents a novel way of automatically extracting User Stories from RE meeting transcripts. The future research section provides suggestions on improving the RSAs and a future view for combining the TUSP with issue-tracking and project management systems.