Evaluating mobile app performance through sentiment analysis with SimCLR and MobileBERT

Ruping Zhang, Aman Ullah, Khair Ullah Khan, Muhammad Nawaz Khan, Rehan Tariq Chohan and Farhan Aadil

PLOS ONE, 2026, vol. 21, issue 6, 1-31

Abstract: Mobile apps have significantly enhanced user engagement and convenience, providing seamless access to services anytime, anywhere. This paper presents a sentiment analysis (SA) model comprising SimCLR (Simple Contrastive Learning of Representations) and MobileBERT to measure the perceived performance of the mobile apps. The reviews of mobile apps are manually labeled as High and Low based on performance parameters (insightfulness, transparency, pertinence, precision, consistency, elaboration, and practical usability insights). These labels are aligned with attitudinal components, i.e., appreciation and judgment from appraisal theory. Appreciation concerns insightfulness, transparency, and pertinence of mobile app features, while judgment pertains to precision, consistency, elaboration, and practical usability. By integrating appraisal theory into this proposed model, a more comprehensive insight into user sentiment is achieved as compared to traditional SA models, which are heavily dependent on a very limited set of sentiment categories, i.e., positive and negative. The accuracy of manual annotation is ensured by the SHAP (SHapley Additive exPlanations) method of Explainable AI (XAI). In this study, SimCLR, a self-supervised learning framework, is employed to improve feature extraction and enhance the model’s ability to generalize across different datasets, while MobileBERT, a lightweight transformer model for sentiment classification, ensures computational efficiency and high performance. The MobileBERT makes the proposed approach scalable, efficient, and highly suitable for real-time SA in resource-limited environments. Knowledge distillation is incorporated to teach a student model that is similar to the teacher model, increasing the robustness of the system. In this work, the Augmentation of data by replacing synonyms in the data using WordNet increases the quantity of the training data to make the model more robust to linguistic variations. The proposed model is optimized using the PCA (Principal Component Analysis) method for dimensionality reduction. The Optuna method is used for hyperparameter optimization, automating the search for the most effective configuration, including learning rate, batch sizes, etc. K-fold stratified cross-validation ensures model robustness. The results demonstrated that the integration of SimCLR, MobileBERT, distillation, data augmentation techniques, and fine-tuning using Optuna provides a highly efficient and accurate SA-based model, with a mean fold accuracy and ROC AUC of 88.63% and 90.91%, respectively. This approach offers a scalable solution for NLP (Natural Language Processing) tasks. The proposed approach will be highly beneficial not only for the developers, yielding a deeper and nuanced understanding of the user feedback to identify specific areas for improvement in app performance, but also to stakeholders like marketers, product managers, and user experience (UX) researchers.

Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0349555 (text/html)
https://journals.plos.org/plosone/article/file?id= ... 49555&type=printable (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:plo:pone00:0349555

DOI: 10.1371/journal.pone.0349555

Access Statistics for this article

More articles in PLOS ONE from Public Library of Science
Bibliographic data for series maintained by plosone ().