A Hybrid Deep Learning Framework for Improving User Review Classification for Usability and Security & Privacy

User reviews on application platforms contain critical feedback regarding usability and security & privacy concerns. Manually analyzing these reviews is inefficient due to large volume. This research presents a hybrid deep learning framework for automatic classification of user reviews into two categories: Usability and Security & Privacy. Three models are implemented and compared: an LSTM baseline, a Hybrid RNN-LSTM (MAUSPC), and a BiLSTM model with self-attention mechanism. The models are trained on a balanced dataset derived from Google Play Store reviews. The BiLSTM with self-attention demonstrates superior classification accuracy by capturing contextual dependencies from both forward and backward directions while focusing on salient text features. A Flask-based web application is developed to deploy the trained models, offering modules for user registration, authentication, review classification, and session management. The framework provides an automated solution for categorizing user feedback.

Keywords: User Review Classification, Deep Learning, LSTM, Hybrid RNN-LSTM, BiLSTM, Self-Attention, Usability, Security and Privacy, Flask, Natural Language Processing