Performance Analysis of Suggestion Mining for Mobile Apps

Abstract Extracting actionable improvement suggestions from user reviews remains a critical challenge in mobile app development. We propose two novel Natural Language Processing (NLP) techniques specifically designed to bridge this gap. The first technique leverages a fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model, while the second employs a lightweight DistilBERT model. Both are trained on domain-specific action verbs and phrases indicative of desired app changes within the mobile app domain. We evaluate the effectiveness of these techniques through a comprehensive case study analyzing a substantial dataset of 8,061 Mobile app reviews. Our findings demonstrate the efficacy of these techniques in capturing valuable user insights that can inform development efforts and ultimately enhance mobile app quality. This, in turn, has the potential to translate into a more user-centric experience, potentially leading to increased user satisfaction and reduced app churn. Further research could explore advanced analysis of extracted suggestions to prioritize user concerns and inform targeted development initiatives.