Elevating Offensive Language Detection: CNN-GRU and BERT for Enhanced Hate Speech Identification

Abstract

Upholding a secure and accepting digital environment is severely hindered by hate speech and inappropriate information on the internet. A novel approach that combines Convolutional Neural Network with GRU and BERT from Transformers proposed for enhancing the identification of offensive content, particularly hate speech. The method utilizes the strengths of both CNN-GRU and BERT models to capture complex linguistic patterns and contextual information present in hate speech. The proposed model first utilizes CNN-GRU to extract local and sequential features from textual data, allowing for effective representation learning of offensive language. Subsequently, BERT, advanced transformer-based model, is employed to capture contextualized representations of the text, thereby enhancing the understanding of detailed linguistic nuances and cultural contexts associated with hate speech. Fine tuning BERT model using hugging face transformer. To execute tests using datasets for hate speech identification that are made accessible to the public and show how well the method works to identify inappropriate content. By assisting with the continuing efforts to prevent the dissemination of hate speech and undesirable language online, the proposed framework promotes a more diverse and secure digital environment. The proposed method is implemented using python. The method achieves 98% competitive performance compared to existing approaches LSTM and RNN, CNN, LSTM and GBAT, showcasing its potential for real-world applications in combating online hate speech. Furthermore, it provides insights into the interpretability of the model's predictions, highlighting key linguistic and contextual factors influencing offensive language detection. The study contributes to advancing hate speech detection by integrating CNN-GRU and BERT models, giving a robust solution for enhancing offensive content identification in online platforms.