BERT-FRIDE: An Efficient Approach for Front-End Issue Detection and Extraction from User Reviews

According to recent research, app Stores like Google Play, Apple AppStore, and Windows Phone Store contain more than 5 million applications. The reviews provided by users on these platforms contain valuable information that can assist developers in enhancing their apps. The large volume of daily app reviews and their noisy nature make it difficult to extract front-end design issues. Accurately classification of reviews is the only viable option for achieving this goal. We proposed the BERT-FRIDE approach to effectively classify user reviews and extract valuable interface design insights. To achieve this objective, we developed a comprehensive dataset by utilizing our groundbreaking zero-shot annotation technique. The dataset contains nine specific categories of interface design issues. Furthermore, we proposed a tailored loss function to address challenges like underperforming classes, assigning higher weights as needed, resulting in an outstanding reported accuracy of 98%. Additionally, we compare our approach with state-of-the-art deep learning models such as XLNet, RoBERTa, and GPT-2, demonstrating superior performance. Our proposed approach has the potential to greatly benefit front-end developers and improve the overall quality of their products.