Answers are grounded in this paper and cite their sources.
In today's mobile app-dominated era, understanding user sentiments is vital for developers, marketers and decision-makers. Traditional sentiment analysis techniques often fall short in capturing the complex, time-dependent nature of user reviews. This project investigates user sentiments in the top 20 mobile applications on Google Play using advanced deep learning approaches—Long Short-Term Memory (LSTM) networks and multilingual BERT (mBERT). LSTM excels at modelling sequential data, capturing evolving user opinions over time, while mBERT leverages multilingual contextual embeddings to deliver deeper sentiment insights. We sourced 200,000 reviews from Kaggle and refined the dataset to 145,404 reviews through preprocessing steps such as contraction expansion, stop word removal and tokenisation. The dataset was divided into training, validation and test sets to ensure optimal performance and generalisability. The LSTM model, trained over 20 epochs, achieved an accuracy of 92.01%, while the mBERT model, trained over five epochs, reached 96.20%. Both significantly outperformed traditional models like Random Forest and CNN. Word cloud analysis highlighted recurring themes in user feedback, enabling more focused app improvements. This research advances sentiment analysis by addressing temporal and contextual challenges, providing a practical framework for enhancing user satisfaction and supporting strategic decision-making in the competitive mobile app landscape.
No takes yet. Share an insight, caveat, or question.
Kollu et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: