Hate Speech Detection on Twitter Using BERT Algorithm

Abstract

Hate speech on one social media platform, Twitter, is uncommon. Users on the Twitter platform can freely obtain, exchange information, and express opinions. This is one of the main factors that a person can be exposed to hate speech on Twitter. Victims who are exposed to hate speech may suffer from mental health disorders because most victims of hate speech are attacked verbally or emotionally. However, the lack of countermeasures against the detection of hate speech on the Twitter social media platform is still rare. In this study, a simulation was carried out using the website, along with testing and analyzing the detection of hate speech. The test is done by inputting a text on the hate speech website, and then the website will do a preprocessing and analyze this text using the BERT algorithm to classify whether the word is hate speech or not. The training results found that the detection of hate speech on Twitter user accounts using the BERT Algorithm has a 78.69% accuracy, a 78.90% precision, a 78.69% recall, and a 78.77% F1 score against the classification of hate speech groups. Thus users will more easily detect hate speech on Twitter by using the hate speech website.