A Two-Stage Data-Driven Framework for App Quality Monitoring: Integrating CNN-BiLSTM and Laney p’ Control Charts

Abstract

Digital public service applications generate large-scale user reviews that provide valuable feedback on service quality, but their unstructured nature makes continuous monitoring challenging. This study proposes a two-stage data-driven framework that integrates FastText-based Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) sentiment classification with Statistical Process Control using the Laney p $'$ control chart and process performance evaluation. The framework was applied to 15,082 Google Play reviews of the Super App Polri collected from August 2023 to August 2025. The proposed model achieved strong sentiment classification performance, with a testing accuracy of 0.9238 and an AUC of 0.9645, indicating its ability to distinguish positive and negative user feedback effectively. The classified sentiment results were then aggregated over time to monitor changes in negative review proportions. The Laney p $'$ chart identified five out-of-control weeks under historical average-based monitoring, while process performance indices indicated that the service process was not yet capable of meeting the specified quality target. Further severity-based analysis of negative reviews during these out-of-control periods showed that most complaints were associated with critical and marginal issues. These findings demonstrate that combining deep learning-based sentiment analysis with statistical quality control provides an interpretable and practical framework for detecting service instability, prioritizing user issues, and supporting continuous improvement in public digital service platforms.