Deep Transform Ensemble Model for Sentiment Analysis

The ensemble method is a technique that has garnered significant attention in recent years, particularly in the field of sentiment analysis. It leverages the strengths of multiple models to enhance overall performance. Although many ensemble methods for sentiment analysis have been proposed, few have incorporated deep learning models. In this study, we propose an ensemble model based on transformers and deep learning to improve sentiment analysis performance. The proposed model comprises the following main components: (i) an embedding layer, which converts input sentences into vector matrices; (ii) a BERT-LSTM-based sentiment classifier, which extracts and learns global and contextual features from the embedding layer; (iii) a BERT-CNN-based sentiment classifier, which extracts and learns local and semantic features from the embedding layer; (iv) an ensemble layer, which combines the extracted features; and (v) an ensemble classifier layer, which classifies the sentiment of the input sentences. The model is evaluated on four benchmark datasets. Experimental results show that it improves sentiment analysis performance by at least 0.02 and up to 0.05.