How to effectively mine app reviews concerning software ecosystem? A survey of review characteristics

Abstract

App reviews in app stores offer valuable insights into many activities in the software ecosystem, e.g., software development, app marketing, security. As app reviews are known to be error-prone, commonly short, dynamic, and hold domain-specific knowledge, we need mining strategies tailored to these characteristics. To help developers or researchers mine reviews more effectively, in this study, we conduct a systematic literature review on app review mining from the perspective of the characteristics. This survey was conducted on 167 papers published between 2012 and 2022 and focuses on three phases in app review mining: (a) the 167 papers were thoroughly examined to extract practices for the collection of app reviews; (b) a detailed list of review characteristics was summarized through a key-point investigation; (c) the survey presents common handling and applications for each review characteristic. Compared with other literature reviews on app review mining, our paper provides insights from the micro perspective. We have noted a growing trend in the analysis of app reviews, with review rating being the most frequently employed review characteristic. We also observed that the Google Play Store stands out as the most commonly used app distribution platform, and simple random sampling prevails as the most popular review sampling strategy compared to stratified sampling and key-point investigation. Additionally, we have identified domain knowledge, textual content, dynamic nature, and sentiment of reviews as the most promising review characteristics for future studies.