Hate Speech Detection Using Expansion Feature Glove with CNN and Bi-LSTM on Twitter

Muchammad Alfi Karom (Telkom University)
Erwin Budi Setiawan (Telkom University)

Jurnal Informatika dan Rekayasa Perangkat Lunak
Vol. 7 No. 1 (2025): Maret

Twitter is one of the biggest social media digital platforms in Indonesia. It serves a medium for readers worldwide and also can be used a means of disseminating information for everyone. However, some people in social media misuse it to spread hate speech against some certain group or communities. Because hate speech it happens everywhere, we need a system to detect hate speech. Sometimes to detect hate speech in Twitter in can be very difficult because lack of context. Needing feature for this problem can make detect hate speech become more easier. Glove is a feature expansion method combine with feature extraction using N-gram and Term Frequency Inverse Document Frequency(TF-IDF) as a method. Data from that it will processed using a hybrid deep learning that combines Convolutional Neural Networks(CNN) dan Bidirectional Long Short-Term Memory(Bi-LSTM). In this study, author obtained 69,484 data related to hate speech. From this study combine feature extraction and feature expansion method has an impact on this research. Best accuracy with all of method is CNN+Bi-LSTM Hybrid method with 91,69% accuracy on top10. Meanwhile best method for Bi-LSTM+CNN method is 91,33% accuracy on top20.