Performance Comparison of Multinomial and Bernoulli Naïve Bayes Algorithms with Laplace Smoothing Optimization in Fake News Classification

In this digital era, there is a lot of information available, one of which is news information. In the dissemination of this news information, there is real news and fake news. This can have a negative impact that can be felt by the public in making decisions based on wrong information and reporting. Therefore, it is important to develop effective methods and algorithms in identifying and classifying fake news. One approach commonly used in fake news classification is the Naïve Bayes algorithm. This algorithm is based on probability theory and could classify texts into fake news or real news categories. In Naïve Bayes there are several variants that are commonly used, namely the Multinomial Naïve Bayes algorithm and the Bernoulli Naïve Bayes algorithm. In this research, the two algorithms with and without Laplace smoothing be compared. The difference in accuracy is 0.003%. The performance of the model using the multinomial naïve Bayes algorithm is superior to Bernoulli naïve Bayes in terms of accuracy if both do not use Laplace smoothing. However, if Laplace smoothing is carried out on both models, the resulting accuracy has the same value and increases. With a model that uses the multinomial Naïve Bayes algorithm, it increases by 0.002%. Thus, multinomial naïve Bayes is better at classifying hoax news than Bernoulli naïve Bayes and Laplace smoothing can increase the accuracy of both models.