A Comparative Analysis of Sentiment Classification Based on Deep and Traditional Ensemble Machine Learning Models

Abstract

The era of the internet has transformed the way people share their thoughts and viewpoints. It is now achieved mostly through blog entries, product review blogs, social networks, and so on. We get immersive media through online networks, where users notify and affect others through the internet. In this research, positive and negative sentiments are used to do the document-level sentiment analysis using deep and traditional ensemble models. In this study, we attempt to evaluate the performances of recent deep learning ensemble models and traditional ensemble models for obtaining the highest accuracy for binary sentiment classification. Three traditional ensemble models (i.e., Voting Ensemble, Bagging Ensemble, and Boosting Ensemble) and three deep learning ensemble layout models (i.e., 7 Layer Convolutional Neural Network (7-L CNN) + Gated Recurrent Unit (GRU), 7-L CNN + GRU + Globe Embedding, and 7-L CNN + Long Short-Term memory (LSTM) + Attention Layer) have been applied in two different datasets to perform the sentiment classification. The deep learning ensemble models perform better than the traditional ensemble models in most cases. In both of the datasets, the deep learning ensemble models namely 7-L CNN + GRU + Globe and 7-L CNN + LSTM + Attention Layer achieve the highest accuracy by securing 94.19% and 96.37% respectively for the product and restaurant review dataset.