Hate Speech and Offensive Content Detection Using Machine Learning

Social media offers global connectivity but also facilitates the spread of hate speeches and lacks offensive language, creating tremendous challenges. Since user-generated content is voluminous, manual moderation using the input is virtually impossible, hence machines' learning (ML) solutions are conceivably essential. The present work assesses five ML models for automatically detecting hate speech, classifying tweets into three classes, namely hate speech, offense language, or neutral speech. Out of 24,783 tweets, XGBoost achieved the highest accuracy, became the best model. For interpretability, feature importance, confusion matrices, and visualization techniques such as word clouds and tweet-length distributions were investigated. While ML models effectively classify the texts, detecting implicit hate speech and multilingual content remains a hurdle. Future work should seek to investigate better models and contextual analysis for safer spaces for interactions on the Internet.