Detection and Classification of Hate Speech on Twitter Using Convolutional Neural Network (CNN)

Hate speech is any form of communication that attacks a person or group based on attributes such as race, religion, ethnicity, gender, sexual orientation, nationality, or other attributes. It can be verbal, written, or nonverbal. Hate speech can be dangerous because it can lead to violence, discrimination, and social exclusion for certain demographics of people. Social media platforms have become the main platforms for spreading hate speech. Social media companies can combat and minimize the spread of hate speech by educating their users about the adverse effects of hate speech and developing a system to detect, identify, and remove contents that contain hate speech. Convolutional neural network (CNN) is an algorithm that can be used to determine whether a tweet is hate speech or not. The proposed method could facilitate the identification and detection of hate speech on social media platforms. Feature extraction is performed using TF-IDF, with FastText used as feature expansion. In this study, there are three test scenarios that were applied, where we included baselines with TF-IDF, FastText implementation, and the best hyperparameter search. The results showed a significant improvement in accuracy, with the third scenario achieving an accuracy of 86.87%, an increase of about 9.84% compared to the results in the first scenario, which got a result of 77.03%, and the second scenario, which got a result of 85.03%.