Multi-Modal Hate Speech Recognition Through Machine Learning

Because of the speedy surge in social media’s expansion, the proliferation of malicious and harmful poses a substantial worry in contemporary society. The identification of hate speech on platforms like Twitter is crucial for various tasks such as controversial event extraction, AI chatterbot creation, content suggestions, and sentiment analysis. Researchers have invested considerable effort in addressing the challenging task of identifying hostile content due to the rise in hate speech and harmful information. The objective is to classify tweets as Hateful, Offensive, or neither. However, this task is highly complex due to the intricate nature of natural language constructs, encompassing different manifestations of animosity directed at various demographics, and the multitude of ways the same meaning can be expressed.Previous research has predominantly relied on manual feature extraction or employed representation-learning techniques followed by linear classifiers. Nevertheless, deep learning methods have recently demonstrated significant accuracy improvements in complex problems across speech, vision, and text applications. In this study, This paper present an idea for automatic classifications of inappropriate language and expressions of hostility using transfer learning models. In this research, leverage classified tweet datasets obtained from Kaggle and conduct experiments. Findings reveal that the multilingual-BERT model and its pre-trained version deliver superior outcomes. Specifically, pre-trained BERT model notably improves classification accuracy of hateful tweets by up to 92% when compared to other algorithms.