Efficient Hate Speech Detection: Evaluating 38 Models from Traditional Methods to Transformers
2025-09-19
Summary
The study evaluates 38 configurations of models for detecting hate speech, focusing on transformer architectures like RoBERTa, deep neural networks, and traditional machine learning methods. Results indicate that transformers, especially RoBERTa, perform best with over 90% accuracy and F1-scores, while traditional methods like CatBoost and SVM are competitive with lower computational requirements.
Why This Matters
Automated hate speech detection is crucial as social media platforms struggle to balance free expression and user safety. This research provides insights into which models are most effective and efficient, aiding in the deployment of scalable, automated systems for content moderation. Understanding the performance of different models helps in selecting the right approach for detecting hate speech efficiently.
How You Can Use This Info
Professionals in content moderation or social media management can use these findings to implement more effective hate speech detection systems. By choosing models like RoBERTa for better accuracy or SVM for efficiency, organizations can tailor their approach based on resource availability and required precision. Additionally, using balanced, raw datasets can enhance model performance without the need for extensive preprocessing.