Performance Evaluation of Some Selected Ensemble Classifiers for Users' Online Book Reviews Based on Sentiment Analysis

Abstract

User reviews on platforms such as Amazon have become a focal point due to their extensive use in sentiment analysis, which provides valuable feedback to the public, private companies, and governments. Analysing these reviews not only contributes to enhancing the quality of products and services but also supports the development of marketing and financial strategies aimed at boosting profitability and customer satisfaction. Even with several models developed for this task, there is room to enhance the processing, classification, and interpretation of user feedback, thereby assisting product managers in refining product quality. This paper introduces ensemble classifiers designed to categorize reviews as positive, negative, or neutral. The study first assesses the performance of widely used ensemble methods, including Adaptive Boosting (AdaBoost), Categorical Boosting (CatBoost), Gradient Boosting (GBM), and Extreme Gradient Boosting (XGBoost). It then evaluates stacking ensemble classifiers, where outputs from Random Forest (RF), Naive Bayes (NB), and Support Vector Machine (SVM) are combined with CatBoost, XGBoost, AdaBoost, and GBM. Using dataset sourced from Amazon, experimental results demonstrate that stacked ensemble classifiers achieve superior accuracy, 88.23%, 86.01%, 84.56%, and 85.12%, compared to traditional ensemble classifiers.