Hate Speech Detection in Indonesian Song Lyric Using BERT

This study was conducted with the aim of detecting hate speech in song lyrics. The background of this study stems from the numerous negative impacts caused by hate speech. We focus on detecting hate speech in song lyrics because music is a popular form of art that can have significant impacts on people’s lives. There have been many previous studies that attempted hate speech detection using various methods. However, one issue still persists, especially when dealing with slang or uncommon sentence structures. According to the previous works, deep learning algorithms, such as using pre-trained BERT model for sentiment analysis and R-CNN as a classifier, have shown promise in addressing these issues. Hence, we decided to utilize the BERT pre-trained model for sentiment analysis and R-CNN as the classifier in their study. The dataset used in this research was collected from the Twitter dataset on Kaggle and also utilized the Spotify API. The study resulted in an precision score of 100% a recall score of 37% and an F1-score of 54% for the hate speech detection. Our result suggests that more relevant data, such as datasets consisting of poetry or sentences with implicit meanings, will be beneficial to enhance hate speech detection in song lyrics.