Automating Mobile App Review User Feedback with Aspect-Based Sentiment Analysis

Vasileios Ballas, Konstantinos Michalakis, Georgios Alexandridis, George Caridakis

Abstract: Effective user feedback is crucial for enhancing the user experience (UX) of mobile applications. However, manually analyzing user reviews can be time-consuming and labour-intensive. This paper investigates the application of state-of-the-art aspect-based sentiment analysis (ABSA) algorithms to automate user review analysis and feedback. We scrape the most relevant Google Play Store user reviews for 6 distinct applications of unrelated categories and we separate them into single sentences. We employ and fine-tune a BERT-based ABSA model – Aspect Sentiment Triplet Extraction by PyABSA – to extract sentiment triplets (aspect, opinion, polarity) from the review sentences. The results demonstrate that ABSA models can effectively capture user feedback by identifying specific aspects and sentiments related to app features and functionalities. Our framework, which utilizes the ABSA model along with filtering methods via Topic Modeling, can automatically extract sentiment triplets and provide additional suggestions and statistics for the app developers. This framework facilitates efficient and comprehensive user feedback collection, enabling developers to make informed decisions for UX improvement.