Sentiment analysis of product reviews based on BERT_BiLSTM model

With the development of the times and the advancement of technology, e-commerce formed by combining shopping and the Internet has become more and more important, and in just a few decades, people's daily lives have become inseparable from online shopping. Using the Internet to shop has become one of the main shopping methods today. In the process of using online shopping, consumer reviews often determine whether people will buy the relevant products and services. At the same time, consumer products also serve as an important indicator that allows companies to understand product defects and user satisfaction. However, due to the popularity of online shopping, the number of consumer reviews is huge and customers must invest a great deal of time and effort to get useful information from them. This research suggests a bidirectional encoder representations from transformers with bidirectional long short-term memory (BERT_BiLSTM) model-based sentiment analysis approach to address the associated issues. Throughout the study, streamlined Amazon reviews obtained on kaggle were used as a dataset with a sample size of 4 million with both positive and negative labels. First, adjust the trained bidirectional encoder representations from transformers (BERT) model, and the classification process was performed using a Bi-directional Long Short-Term Memory, resulting in a validation set with a stable accuracy of 0.86. The outcomes of the experiment demonstrate that the BERT_BiLSTM model is significantly improved in accuracy and f1 scores in contrast to the Bidirectional Encoder Representations from Transformers with Recurrent Neural Network and BERT models, and when compared to the Bidirectional Encoder Representations from Transformers with Long Short-Term Memory model, accuracy and f1 scores consistently increase.