Attributes that Predict which Features to Fix

Abstract

Requirements engineering is assessed as the most important phase of the software development process. This process is especially challenging for app developers, who tend to gather crowd-based feedback after releasing their apps. This feedback is often voluminous, posing prioritization challenges for developers identifying features to fix or add. While previous work has identified frequently mentioned features, and some effort has been dedicated towards providing various prioritization and classification techniques, these do not quite address the prioritization challenge faced by app developers given voluminous app reviews. In fact, there is also need to assess the scale of app reviews' usefulness. We use content analysis and regression to contribute towards this cause by exploring the usefulness of app reviews, and the attributes that predict which app features to fix, respectively. Our outcomes show that reviews tended to either provide information of little value (i.e., no actionable information) or highlighted problems that may directly affect the functionality of app features. For two different apps, we also observe that features that were mentioned the most (the feature frequency attribute) in lower ranked reviews provided by users had the strongest predictive power for identifying severely broken features (as perceived by a developer). However, the ordering did not match with the frequency with which reports were made by users. There were also variances in the attributes that predict which features to fix, for the reviews of different apps. Review mining and prioritization challenges remain given variances in app reviews' content and structure. These findings also point to the need to redesign app review interfaces to consider how reviews are captured.