Consumer Sentiment Analysis in the Market through Online Product Reviews Using BERT-BiLSTM on Tiki Data

Abstract

Nowadays, with the rapid growth of social media and the surge in e-commerce, customers' expression of opinions through online product reviews has become an indispensable part of the landscape. Sentiment analysis plays a pivotal role in automatically extracting subjective information, particularly customers' emotions and opinions, from these reviews. Recognizing the importance of this, this article introduces a comprehensive system that starts with data collection from the Tiki e-commerce website. Subsequently, it employs the Bidirectional Encoder Representations from Transformers (BERT-BiLSTM), a predictive model that fuses Bidirectional Encoder Representations from Transformers (BERT) and Bidirectional Long Short-Term Memory (BiLSTM). This model aims to forecast emotional trends within user product reviews accurately. The results indicate that the BERT-BiLSTM model achieves remarkable accuracy, recall, precision, and an impressive F1 score of 93.14%, 94.46%, 93.56%, and 93.83%, respectively. Consequently, the findings of this study provide a precise means of predicting consumers' emotional trends, particularly concerning specific product lines.