Enhancing the accuracy of transformer-based embeddings for sentiment analysis in social big data

Social media have opened a venue for online users to post and share their opinions in different life aspects, which leads to big data. As a result, sentiment analysis has become a fast-growing field of research in Natural Language Processing (NLP) owing to its central role in analysing the public's opinion in many areas, including advertising, business and marketing. This study proposes a transformer-based approach, which integrates contextualised words with Part-of-Speech (POS) embedding. Then, the enhanced word vector is forwarded to a hybrid deep learning architecture combining a Convolutional Neural Network (CNN) and a Bidirectional Long Short-Term-Memory (BiLSTM) to discover the post's sentiment. Extensive experiments on four review data sets from diverse domains demonstrate that the proposed method outperforms other machine learning approaches in terms of accuracy.