Identification and classification of requirements from app user reviews

Abstract

Review function, as a feedback mechanism from users to developers and vendors, is provided by most APP distribution platforms that allow users to rate and comment an APP after using it. User reviews are recognized as a valuable source to improve APPs and increase the value for users. With the sharp increase in the amount of user reviews, how to effectively and efficiently analyze the user reviews and identify potential and critical user needs from them to improve the APPs becomes a challenge. In this paper, we propose an approach to automatically identify requirements information and further classify them into functional and non-functional requirements from user reviews, using a combination of information retrieval technique (TF-IDF) and NLP technique (regular expression) with human intervention in keywords selection for requirements identification and classification. We validated the proposed approach with the user reviews collected from a popular APP iBooks in English App Store, and further investigated the cost and return of our approach: how the size of sample reviews for keywords selection (cost) affects the classification results in precision, recall, and F-measure (return). The results show that when setting an appropriate size of sample reviews, our approach receives a relatively stable precision, recall, and F-measure of requirements classification, in particular for non-functional requirements, which is meaningful and practical for APP developers to elicit requirements from user reviews.