Detection of Indonesian Hate Speech on Twitter Using Hybrid CNN-RNN
Slamet Riyadi, Annisa Divayu Andriyani, Ahmad Musthafa Masyhur, Cahya Damarjati, Mahmud Iwan Solihin · 2023
This research focuses on detecting hate speech on Twitter in Indonesia. The Indonesian government has enacted legislation to address hate speech issues, but certain provisions within the Information and Electronic Transactions Law (UU ITE) require revaluation due to open interpretations. To improve hate speech detection, the researchers employ a combination of deep learning methods, namely CNN and RNN, to analyze offensive language on Indonesian Twitter. The Hybrid CNN-RNN model leverages CNN's ability to extract local features and RNN's capability to model sequential context, ensuring accurate representation and addressing contextual dependencies and variable dimensions. The research aims to enhance the accuracy of hate speech detection by combining the Hybrid CNN-RNN model with Word2Vec word embedding. Previous studies achieved an accuracy of 69.1% using the same method, motivating the need for improvement. The proposed method demonstrates significant progress, with an accuracy of 0.863. The conclusion drawn suggests that the Hybrid CNN- RNN model with Word2Vec embedding is the most effective classification model for detecting hate speech on Indonesian Twitter. Future research recommendations include utilizing larger datasets and implementing enhanced preprocessing techniques to enhance accuracy further.