User Reviews Classification in Play Store Applications Using Deep Learning: An Empirical Study

Seyyed Mehrdad Razavi Dehkordi, Hamid Rastegari, Akbar Nabiollahi-Najafabadi, Taghi Javdani Gandomani


Since developing mobile applications, users usually express their requirements as reviews under applications residing in the Google Play Store or Apple AppStore. Some methods were proposed for automatically classifying user reviews, most of which employ old methods and databases or provide poor accuracy. In this article, a model named DARCLSTM is proposed to improve the process of user reviews classification. In the proposed model, the New Kaggle dataset containing 51,000 reviews related to the year 2021 is given to the deep learning system with LSTM architecture to train the model after pre-processing, removing noisy data and data cleaning along with the reviews lables. Then the trained model is used to classify reviews into three groups bug reports, feature requests, and information_giving while using an app. Then, the model is compared to other methods (other proposed models using machine learning or deep learning), indicating the outperformance of proposed modelĀ  against previous studies. The F-measure (11%) and accuracy (97%) parameters have significantly improved in the proposed model.


Mobile application, User reviews, google play store, classification, deep learning, machine learning;