Developing Hybrid Model for Analyzing Sentiment of Textual Data for Amazon Product Reviews Using Deep Learning

Abstract

The article focuses on sentiment analysis, a popular research area within Natural Language Processing, driven by the growth of social networks and e-commerce websites. As usual, there are five rates used to determine the rating of each review based on customer feedback for any product. The study aims to reduce the number of rating categories on the Amazon platform from five to three classes: high rate, middle rate, and low rate. The paper proposes a Bidirectional LSTM model for sentiment analysis, utilizing deep learning techniques to accurately classify users' thoughts and emotions. Accordingly, we have developed a new model called DCNN to perform the classification process for customer reviews ratings, and our proposed model has been trained on a large dataset of reviews. The study utilized a dataset of Amazon product reviews for analysis. To enhance the data for our model, we conducted several data preprocessing steps. We transformed the dataset from an imbalanced dataset to a balanced one. As a result, our model achieved a significantly improved level of accuracy compared to previous models.