Enhancement of multinomial naïve bayes algorithm applied to sentiment analysis on reddit posts

ABSTRACT: Sentiment classification plays a crucial role in analyzing textual data, but traditional Multinomial Naïve Bayes (MNB) often struggles with class imbalance, high dimensionality, and zero probability issues. This study indented to enhance the performance of MNB by addressing these limitations through random oversampling, chi-square feature selection, and Laplace smoothing. The research focused on sentiment analysis of Reddit posts and conducted experiments on nine datasets, with five datasets containing 1000 data points each and four datasets containing 6000 datapoints each. The enhanced algorithm demonstrated significant improvements in classification accuracy and execution time. For datasets with 1000 data points, the enhanced MNB achieved an average accuracy improvement of 69.14% compared to the traditional approach, while datasets with 6000 data points showed an average accuracy improvement of 50.29%. Additionally, chi-square features, resulting in an average execution time improvement of 35.31% for the smaller datasets and 23.00% for the larger datasets. Laplace smoothing successfully addressed the zero-probability issue, ensuring more robust probability estimates in classification. The findings of this study confirm that addressing these algorithmic limitations significantly enhances the effectiveness and efficiency of sentiment classification using MNB. Future research may explore the application of the enhanced algorithm in other domains, such as customer feedback analysis, political sentiment detection, and social media monitoring, to further validate its adaptability and performance

Filipiniana