Sentiment Analysis Using Self-Adaptive Stacking Ensemble Method for Classification

Abstract

The primary purpose of sentiment analysis is to classify the polarity of the data, such as whether the data should be positive, negative, or neutral. Most sentiment analyses used single classifiers, but they do not provide an accurate polarity. There should also be drawbacks, like a lack of keywords, high dimensional space, etc. This paper used the polarized word embedding technique and Remora Optimization algorithm for distance ranking; then, the classification is done by both machine learning and deep learning classifiers that are integrated using the self-adaptive stacking ensemble method to select the finest base classifier and hyper-parameters of base classifiers with the use of the genetic algorithm. Then, the model is trained and tested employing four datasets utilizing cross-validation, and the performance is calculated using recall, accuracy, precision, F1 score, and AUC that is compared using four state-of-the-art models. The comparison shows that the proposed method provides the most accurate predicted value with the highest accuracy of 99.3%.