Sentiment Analysis of the Consumer Review Text Based on BERT-BiLSTM in a Social Media Environment

Introduction

As information technology and mobile Internet are being developed and globalized rapidly, the Internet has entered thousands of households and become a part of people’s lives; moreover, online shopping has the greatest impact on such users (Yuan et al., 2020). Studies have shown that consumers like to comment on purchased goods when shopping online (Demotte et al., 2020; Yadav & Vishwakarma, 2019). The vast number of Internet users has moved from simple information recipients to main publishers of online content. Furthermore, such simple consumers have gradually begun to increase the behavior of “publishing purchase comments” where a large number of exponential growths of short text data appear on various websites, and the comments on e-commerce websites have the characteristics of a huge number that is usually mixed with praise and criticism (Alharbi et al., 2020; Dellal-Hedjazi & Alimazighi, 2022; Elfaik & Nfaoui, 2021). As consumers cannot see the real object or make a direct judgment on its quality during online shopping, they are more willing to understand the details of the goods through the objective information of the buyer comments than the subjective description of the business itself (Abdullaha et al., 2021; Chandrasekaran et al., 2022). Furthermore, sellers of e-commerce platforms can use these comments to understand people's reactions toward the goods and understand their problems in order to formulate reasonable sales strategies (Prabha & Srikanth, 2019; Zhou et al., 2020).

Analyzing such large amounts of data and extracting valuable information from them manually is a time-consuming, difficult, and even impossible operation (Li et al., 2019; Yadav & Vishwakarma, 2020). Therefore, the proposed method has a significant social role in helping people to gain and analyze the important business and social value information contained in the collected data by automatically acquiring and processing a large amount of text data (Wang et al., 2020; Seo et al., 2020; Shi et al., 2019).

The use of artificial intelligence (AI) and related technologies in natural language processing as well as the in-depth study of natural language processing with related technologies, used to solve issues in the sentiment analysis of online comment texts, are currently hot research topics both domestically and internationally (Colon-Ruiz & Segura-Bedmar, 2020; Kumari et al., 2021). Therefore, using the BERT pre-training language model to process natural language and applying the BiLSTM method to train all the data and fully extract the features contained in the text are the main phases of this work. The combination of BERT model and BiLSTM is applied to the sentiment analysis. Compared with traditional sentiment analysis methods, the innovations of the proposed method lie in the following:

• 1.

  While using the BERT pre-training language model to process different downstream tasks, only fine-tuning is needed, which reduces the amount of calculation.

• 2.

  BiLSTM can comprehensively get the relevant information in the context and fully extract the features contained in the text.

• 3.

  Taking the word vector generated by BERT as the input sequence for BiLSTM, it can better combine the context to classify the sentiment and improve the accuracy.