Review Rating Predictions Using Improved Deep Learning Architecture

Understanding consumer sentiment has always been an important aspect of decision-making processes in e-commerce and service industries, therefore review rating prediction plays an important role in assessing it. Users' experiences are enhanced through accurate models that predict ratings reliably from reviews and feedbacks about products and services. This paper presents a novel hybrid deep learning-based model using 1D Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU) to address the task of review rating prediction. By utilizing Global Vectors for Word Representation (GloVe) embeddings and Term Frequency-Inverse Document Frequency (TF-IDF), our model can effectively capture both local patterns and sequential relationships of the text data. Experimental results on the Amazon 5-core software review dataset demonstrate that the proposed model attains 84.31% accuracy with TF- IDF, performing better than deep learning-based models. The proposed model demonstrates high recall, precision, and F1 scores, indicating the suitability of the proposed model for rating prediction. In addition, confidence interval tests justify the statistical reliability of the predictions achieved by the proposed model in producing a very low error rate and small error margin. This demonstrates the ability of the model to enhance the prediction of review ratings.