Classification Cyber Harassment on Twitter using Multinomial Naïve Bayes

Abstract

Multinomial Naïve Bayes is a classification method in Naïve Bayes Classifier based on Bayes Theorem and multinomial distribution. This method works optimally in the multiclass classification of text data. Furthermore, it calculates the probability of occurrence of each word by multiplying the class prior probability by the likelihood value of the occurrence of each word in each class. The phenomenon of Cyber Harassment is defined as the behavior of utilizing technology to harm or humiliate people, which has four types of behavior, namely Physical Threats, Purposeful Embarrassment, Racist, and Sexual Harassment. The number of Cyber Harassment cases always increases every year even though the government has made policies to deal with Cyber Harassment cases. The study aims to classify results accurately regarding the types of Cyber Harassment on Twitter using the Multinomial Naïve Bayes method. The classification results obtained are 20 tweets classified as Physical Threats, 10 tweets classified as Purposeful Embarrassment, 25 tweets classified as Racist, and 22 tweets classified as Sexual Harassment. The accuracy of classification of types of Cyber Harassment on Twitter social media using Multinomial Naïve Bayes is 77%, and the results of the model performance test with K-fold cross-validation is 76.21%, showing that the Multinomial Naïve Bayes method can classify the types of Cyber Harassment on Twitter social media is well effective.