Sentiment Analysis of Social Media Posts using BERT: A Case Study on Consumer Products

Abstract

Sentiment Analysis is important to analyze consumer feedback especially in the social media where large amounts of data are being created every day. The paper explores how BERT (Bidirectional Encoder Representations from Transformers) a state of the art deep learning model may be used to analyze the sentiment of social media posts on consumer products. We use BERT on a customer review case study on Twitter, Instagram, and Facebook of a well-known consumer product. The analysis categorizes the sentiment into three namely, positive, neutral and negative giving an insight about the consumer sentiment. This paper presents a research proposal to a new sentiment analysis system with BERT networks to classify consumer review texts in the e- commerce Big Data setting. First, the author applies the pre-trained model of BERT to unlabeled text allowing the retrieval of more informative and contextually informed word embeddings. These embeddings are feature vectors that contain more textual features. The fusion of BERT enables us to narrow down the feature vector representation and the sentiment classification becomes more precise. Build a sentiment analysis model based on the integrated BERT model which is aimed at enhancing the effectiveness of sentiment classification by combining contextual features. The suggested approach is evaluated against three other conventional sentiment analysis approaches based on the identical dataset. The experimental data prove that BERT-based approach is better than the others with the precision, recall, and F1-Measure values of 92.64, 90.32, and 91.46, respectively. These findings suggest that BERT has been effective in deriving subtle contextual data and enhances the accuracy of sentiment analysis in consumer reviews in the e-commerce industry.