Hierarchical Multi-label Classification to Identify Hate Speech and Abusive Language on Indonesian Twitter

Abstract

Hate speech is one type of speech whose spread is banned in public spaces such as social media. Twitter is one of the social media used by some people to broadcast hate speech. The hate speech can be specified based on the target, category, and level. This paper discusses multi-label text classification using a hierarchical approach to identify targets, groups, and levels of speech hate on Indonesian-language Twitter. Identification is completed using classification algorithms such as the Random Forest Decision Tree (RFDT), Nave Bayes (NB), and Support Vector Machine (SVM). The feature extraction used for classification is the term frequency feature such as word n-gram and character n-gram. This research conducted five scenarios with different label hierarchy to find the highest accuracy that can possibly be reached by hierarchical classification. The experimental results show that the hierarchical approach with the SVM algorithm and word uni-gram feature has an accuracy of 68.43%. It proved that the hierarchical algorithm can increase data transformation or flat approach.