Classifying User Apps Review for Software Evolution: A Preliminary Experiment

Abstract

Application Store is a platform where users can download several applications and games. Users also can provide comments about related applications. These comments made as evaluation material for developers, who have not yet developed applications in the future. In previous studies, an application user assessment has been carried out based on existing taxonomies such as feature requests, information provision, information retrieval, and problem discovery by using Natural Language Processing (NLP), Text Analysis (TA) and Sentiment Analysis (SA). In this study, we propose a model using Topic Modelling (TM) and Minority Synthetic Over-Sampling Technique (SMOTE) to improve classification results. Making user reviews that previously ignored can be taken into consideration for developers in conducting software development. Topic modelling will generate list of topics that representing each review and SMOTE method can overcome the amount of imbalanced data on several tables. We also combine methods TA + NLP + SA, TA + NLP + SA + TM, and TA + NLP + SA + TM + SMOTE with J48 classifier....