Analisis Perbandingan Tingkat Performa Algoritma SVM, Random Forest, dan Naïve Bayes untuk Klasifikasi Cyberbullying pada Media Sosial

Abstract

In January 2022, the number of Internet users in the world has reached 4,95 billion with an average of activity of 135 to 193 minutes per day. Technological advances in information gathering and communication are not in line with the improvements in people's behavior on social media. It is recorded that most of cyberbullying incidents in 2017 originate from social media. Social media are media technologies that facilitate interaction between people on the Internet. The most used social media in the world are Youtube, Instagram, Snapchat, Whatsapp, dan Twitter. There is a static data indicating that 54% of participants in The Annual Bullying Survey have experienced cyberbullying. For this research, a sentiment analysis was performed on a collection of texts from several social media platforms around the world. There are about 46000 different texts with an approximately 8000 text for each category, namely age cyberbullying, ethnicity cyberbullying, gender cyberbullying, religion cyberbullying, other type of cyberbullying and not cyberbullying and approximately 1000 text consist word “fuck”. Sentiment analysis is the process of classifying sentiments in text, whether or not the text contains cyberbullying emotions. This research classifies the type of cyberbullying using the TF-IDF (Term Inversion Frequency Document) function and 3 models namely SVM (Support Vector Machine), RF (Random Forest) and Naive Bayes. Result highlight that SVM and Random Forest performed the best and achieved a precision 82%, recall 83%, accuracy 83% and precision 83%, recall 82%, accuracy 82% using evaluation matrix.