An integrated approach to text sentiment analysis using BERT-iBiLSTM and dual Moth Flame Optimization

Abstract

The massive growth in online content on various social media sites has resulted from the increased use of the internet over the last few decades, covering a wide range of subjects, including situations, events, products, and services. Sentiment analysis (SA) is an algorithmic recognition and classification of the opinions expressed within a text to determine the positive, negative, or neutral disposition of the person reviewing it regarding a certain subject. The current systems have constraints, including a lack of contextual knowledge, substandard treatment of sequential relationships, and a suboptimal selection of features. To enhance the efficiency of these models, this research will propose an integrated solution for sentiment analysis of textual data that includes a novel BERT-iBiLSTM model, which combines bidirectional encoder representations of transformers with enhanced bidirectional long short-term memory for feature extraction. Moreover, the research utilises a dual moth flame optimisation approach to select the optimal features. Finally, an ensemble classifier is employed to identify positive, negative, or neutral sentiments.  The proposed model yields an accuracy of 98.77% on the restaurant review dataset, 96.82% on Amazon and 98.22% on the IMDB datasets, which is similarly satisfactory compared with other state-of-the-art methods. The proposed model can be applied in the fields of social media monitoring, market research, and consumer feedback analysis to acquire significant insights on public sentiment and user contentment.