Visualizing User Story Requirements at Multiple Granularity Levels via Semantic Relatedness

The majority of practitioners express software requirements using natural text notations such as user stories. Despite the readability of text, it is hard for people to build an accurate mental image of the most relevant entities and relationships. Even converting requirements to conceptual models is not sufficient: as the number of requirements and concepts grows, obtaining a holistic view of the requirements becomes increasingly difficult and, eventually, practically impossible. In this paper, we introduce and experiment with a novel, automated method for visualizing requirements—by showing the concepts the text references and their relationships—at different levels of granularity. We build on two pillars: (i) clustering techniques for grouping elements into coherent sets so that a simplified overview of the concepts can be created, and (ii) state-of-the-art, corpus-based semantic relatedness algorithms between words to measure the extent to which two concepts are related. We build a proof-of-concept tool and evaluate our approach by applying it to requirements from four real-world data sets.