Naïve-Bayes family for sentiment analysis during COVID-19 pandemic and classification tweets

Abstract

This paper proposes a system to analyze the sentiments of tweeters. It is to build an accurate model to detect different emotions in a tweet. The analysis takes place through several stages (i.e., pre-processing, feature extraction, and training more than one machine learning (ML)). Naïve Bayes, Multinomial Naïve Bayes and Bernoulli Naïve Bayes were selected as supervised machine learning for sentiment analysis using a dataset of 3,057 tweets with users ranging from fear to happiness, anger, and sadness because this method is suitable for solving a problem of this type. This system was also applied to another dataset of 10,000 Tweets (5,000 positive and 5,000 negatives). This approach, consisting of three Naïve Bayes classification models, was applied to two datasets to analyze the sentiment used in them and classify each category separately. The Multinomial Naïve Bayes model outperformed the other models Where it achieved an accuracy of (91.6%) when applied to the first dataset and accuracy (87.6%) when applied to the second dataset. The researchers aim to continue this research with larger data by using other methods of sentiment analysis to predict users' thoughts about COVID-19 or any other problem and to obtain higher accuracy for the models used.