Combatting Online Hate: A Study on Detecting and Preventing Hate Speech Across Social Media Platforms

Abstract

Social media now has a vital function in our daily lives with the advancement of technology and easy access to the internet. Especially on Twitter, thousands of tweets are shared every day. Alongside constructive comments, numerous tweets containing hate speech are also shared. Many comments contain hate speech on Twitter and other online platforms such as e-commerce sites, forums, news sites, and other social media platforms. The widespread use of hate speech in online environments leads to the marginalization of different groups and thoughts that diverge from the established ideology in societies, resulting in prejudice and discrimination from the community. This study is conducted to prevent the negative consequences caused by hate speech. In the study, two datasets obtained from Kaggle were used. The first dataset comprises data obtained from Twitter, while the second dataset comprises comments made on other social media platforms. The Random Oversampling technique was employed to adjust the dataset imbalance. Random Forest (RF), AdaBoost, Extreme Gradient Boosting (XGB), CatBoost, Bidirectional Long Short-Term Memory (Bi-LSTM), Long Short-Term Memory (LSTM), and Gated Recurrent Unit Networks (GRU) were utilized. The accuracy rates, confusion matrices, precision, recall, F1-scores, and accuracy/loss plots obtained from algorithms were compared. The Bi-LSTM achieved the highest performance with a rate of 94.58%.