Sentiment analysis on store reviews with deep learning method: The example of Amazon

In today's world, e-commerce and modern logistics are rapidly developing. One distinguishing feature of the internet age is that more people are converting their shopping preferences into online product purchases through e-commerce platforms. However, online shopping poses many challenges due to inconsistencies between product descriptions and the physical appearance of the product, as it allows customers to navigate only virtual products. To ensure that the product matches its descriptions, customers use product reviews as an important reference. Therefore, product reviews should be a fundamental index for evaluating products. This study presents a case study focusing on Amazon product reviews that cover both textual comments and star ratings ranging from one to five. It evaluates the consistency between textual comments and given ratings, addressing the challenge by considering a range of classical algorithms such as Support Vector Machine, Decision Tree (DT), and K-Nearest Neighbor, as well as popular deep learning techniques like Long Short-Term Memory (LSTM). This study includes various model comparisons, such as comparing the performance of models and examining the accuracy trend based on the number of hidden layers in deep learning models. In the preprocessing section of the study, the dataset is divided into three parts: training, validation, and testing. The result of the study shows that the LSTM model achieved an impressive accuracy rate of 98%, while the DT model displayed the lowest accuracy at 77.8%.