Prioritisation of requests, bugs and enhancements pertaining to apps for remedial actions. Towards solving the problem of which app concerns to address initially for app developers

Abstract

Useful app reviews contain information related to the bugs reported by the app’s end-users along with the requests or enhancements (i.e., suggestions for improvement) pertaining to the app. App developers expend exhaustive manual efforts towards the identification of numerous useful reviews from a vast pool of reviews and converting such useful reviews into actionable knowledge by means of prioritisation. By doing so, app developers can resolve the critical bugs and simultaneously address the prominent requests or enhancements in short intervals of apps’ maintenance and evolution cycles. That said, the manual efforts towards the identification and prioritisation of useful reviews have limitations. The most common limitations are: high cognitive load required to perform manual analysis, lack of scalability associated with limited human resources to process voluminous reviews, extensive time requirements and error-proneness related to the manual efforts. While prior work from the app domain have proposed prioritisation approaches to convert reviews pertaining to an app into actionable knowledge, these studies have limitations and lack benchmarking of the prioritisation performance. Thus, the problem to prioritise numerous useful reviews still persists. In this study, initially, we conducted a systematic mapping study of the requirements prioritisation domain to explore the knowledge on prioritisation that exists and seek inspiration from the eminent empirical studies to solve the problem related to the prioritisation of numerous useful reviews. Findings of the systematic mapping study inspired us to develop automated approaches for filtering useful reviews, and then to facilitate their subsequent prioritisation. To filter useful reviews, this work developed six variants of the Multinomial Naïve Bayes method. Next, to prioritise the order in which useful reviews should be addressed, we proposed a group-based prioritisation method which initially classified the useful reviews into specific groups using an automatically generated taxonomy, and later prioritised these reviews using a multi-criteria heuristic function. Subsequently, we developed an individual prioritisation method that directly prioritised the useful reviews after filtering using the same multi-criteria heuristic function. Some of the findings of the conducted systematic mapping study not only provided the necessary inspiration towards the development of automated filtering and prioritisation approaches but also revealed crucial dimensions such as accuracy and time that could be utilised to benchmark the performance of a prioritisation method. With regards to the proposed automated filtering approach, we observed that the performance of the Multinomial Naïve Bayes variants varied based on their algorithmic structure and the nature of labelled reviews (i.e., balanced or imbalanced) that were made available for training purposes. The outcome related to the automated taxonomy generation approach for classifying useful review into specific groups showed a substantial match with the manual taxonomy generated from domain knowledge. Finally, we validated the performance of the group-based prioritisation and individual prioritisation methods, where we found that the performance of the individual prioritisation method was superior to that of the group-based prioritisation method when outcomes were assessed for the accuracy and time dimensions. In addition, we performed a full-scale evaluation of the individual prioritisation method which showed promising results. Given the outcomes, it is anticipated that our individual prioritisation method could assist app developers in filtering and prioritising numerous useful reviews to support app maintenance and evolution cycles. Beyond app reviews, the utility of our proposed prioritisation solution can be evaluated on software repositories tracking bugs and requests such as Jira, GitHub and so on.