Extracting and Minimizing Relations for Enhanced Coverage of User Stories

User stories are the first choice of practitioners for expressing requirements in agile projects. ‎Semi-structured natural language (NL) user stories, though easy to read and write, cannot ‎accurately represent a problem domain holistically. The conceptual model, constituting ‎relations (e.g., Teacher teaches Students) among key concepts in a domain, comes in handy to ‎serve the purpose. Several NLP-based approaches exist that extract these relations ‎automatically from user stories for conceptual modeling. These approaches do not make ‎optimum use of NLP capabilities, and consequently, inefficiency and incompleteness in the ‎extracted models are observable. To deal with these issues, we propose an approach for ‎relation extraction using the Open Information Extraction (OpenIE) NLP technique. The ‎OpenIE facilitates our process by automatically extracting relation triples (subject, relation, ‎object). The primary extraction results in a multitude of relations that we reduce into a ‎minimal set by applying a two-step reduction process. The minimal set of relations helps us ‎achieve efficiency as we plate up necessary minimum relations for further processing in ‎software development. The expanded coverage of the input user story by the minimal relation ‎set indicates a high degree of completeness, as our minimal set of relations represents most of ‎the domain knowledge hemmed in by the user story. We devise two metrics, Relation Set ‎Reduction (RSR) and Relation Set Coverage (RSC), to evaluate our approach. The evaluation ‎of two extensive user story datasets shows promising results as we are able to achieve a ‎‎65.91% reduction (RSR) in relations, with 86.45% coverage (RSC) of the user stories‎.