Indonesia Hate Speech Detection using Deep Learning

Hate speech brings negative impacts not only to the target victim, but also to the listener. The spread of hate speech can be done not only through the social media postings, but also through the video, campaign or speech. In this research, we develop models to detect hate speech in Indonesian Language from input text and speech by using deep learning approach. We utilized both textual and acoustic features and compare their accuracies. Experiments result showed that hate speech detection using only textual features is better than that of using acoustic features and both of combined features model. The best model using textual feature obtained F1-score 87.98% which is higher than the model of using acoustic feature only (F1-score 82.5%), and the model of using acoustic and lexical features (F1-score 86.98%).