Mining User Reviews for Software Requirements of A New Mobile Banking Application

Abstract

Migration to the new system or application is very challenging, especially if the users have to adapt to a new application that is implemented with direct conversion technique. It triggers many user reactions, one of them is their opinions and rate about the application in play store (Google Play Store for example). Application reviews can be used to elicit user requirements or to verify requirements. This paper demonstrated the result of mining application reviews to support software requirements elicitation. It motivated by research area natural language processing (NLP) for requirement engineering (RE). Training and testing conducted to a dataset contains about 1200 application reviews of a new mobile banking application by classifying them into two classes (req and other) using Multinomial Naïve Bayes algorithm. Req is for opinions that contain requirement such as feature addition or user interface (UI) request while other is label for opinions/reviews contain non-requirements. The classification performance measured are accuracy score 0,8220 and one of class that has higher classifier performance is “other” class with value precision 0.83, recall 0.94 and F1 0.99. Even though, the result is not optimal yet, especially for “req” class, this research already implemented all categories of NLP technologies such as NLP techniques, NLP tools, and NLP resources.