Classification of Product Review Sentiment Using Naive Bayes and Support Vector Machine Algorithms

Abstract

This review synthesizes research on "Theoretical comparison of Naïve Bayes and Support Vector Machine for sentiment classification of product reviews" to address inconsistencies in algorithm performance and applicability across diverse datasets. The review aimed to evaluate theoretical foundations and practical implementations of both algorithms, benchmark classification metrics, analyze factors influencing performance, assess handling of neutral sentiments, and examine ensemble model efficacy. A systematic analysis of studies from Southeast Asia and related regions was conducted, focusing on supervised learning approaches with varied preprocessing and evaluation metrics. Findings indicate that Support Vector Machine generally achieves higher accuracy, precision, and recall across balanced and large datasets, while Naïve Bayes offers superior computational efficiency and recall in specific contexts. Preprocessing techniques and dataset characteristics significantly affect both algorithms’ robustness, with Support Vector Machine demonstrating greater adaptability to data variability and neutral sentiment classification. Hybrid and ensemble models combining Naïve Bayes and Support Vector Machine consistently improve classification accuracy and robustness but incur higher computational costs and remain underexplored. These results underscore the necessity of context-specific algorithm selection and optimization in sentiment analysis. The review highlights theoretical and practical implications for deploying machine learning classifiers in product review sentiment tasks, emphasizing the balance between accuracy, efficiency, and scalability within resource and data constraints.